Electrical Design and Operation of Sustainable Business Parks

Electrical design and operation of sustainable business parks

Simon De Clercq

Supervisor: Prof. dr. ir. Lieven Vandevelde Counsellor: Dr. Brecht Zwaenepoel

Master's dissertation submitted in order to obtain the academic degree of Master of Science in Electromechanical Engineering

Department of Electrical Energy, Metals, Mechanical Constructions & Systems Chair: Prof. dr. ir. Luc Dupré Faculty of Engineering and Architecture Academic year 2016-2017

Electrical design and operation of sustainable business parks

Simon De Clercq

Supervisor: Prof. dr. ir. Lieven Vandevelde Counsellor: Dr. Brecht Zwaenepoel

Master's dissertation submitted in order to obtain the academic degree of Master of Science in Electromechanical Engineering

Department of Electrical Energy, Metals, Mechanical Constructions & Systems Chair: Prof. dr. ir. Luc Dupré Faculty of Engineering and Architecture Academic year 2016-2017 The author gives permission to make this master dissertation available for consultation and to copy parts of this master dissertation for personal use. In the case of any other use, the copyright terms have to be respected, in particular with regard to the obligation to state expressly the source when quoting results from this master dissertation.

Ghent, June 2017

The promotor The advisor The author

Prof. dr. ir. L. Vandevelde Dr. B. Zwaenepoel Simon De Clercq Preface

It is a pleasure to be able to ﬁnish my engineering studies by writing this dissertation thesis. This work, and the learning process that preceded it, have allowed me to explore the multidisciplinary world of industrial symbiosis and the integration of renewable power sources. It has taught me to enjoy the freedom of independent research and gave me the opportunity to broaden my vision on renewable energy.

I would like to thank my promotors professor Vandevelde and Brecht Zwaenepoel for their support and guidance over the course of this work. I also particularly thank Hendrik Ver- meersch for his input throughout the year; our conversations on the Belgian energy sector were a pleasure. I would like to gratefully acknowledge Christof Deckmyn for his explanation on genetic algorithms.

I want to thank Mieke Gevaerts at SOGent for introducing me to Eiland Zwijnaarde and showing me a glimpse of what urban development is.

To everyone who took their time to review my writing, Brecht, Hendrik and Samie: thank you very much.

Finally, I would like to thank everyone who was there for me, my family, friends, fellow students, house mates, . . . Thank you for making this year so enjoyable. Summary

Electrical design and operation of sustainable business parks Simon De Clercq

Supervisor: Prof. dr. ir. Lieven Vandevelde Counsellor: Dr. Brecht Zwaenepoel

Faculty of Engineering and Architecture Ghent University Academic Year 2016-2017

Abstract - In this study a techno-economic optimisation is carried out on the design and operation of energy systems on newly-developed industrial parks. An optimal sizing model is constructed that uses a genetic algorithm to find the optimal system configuration of an on- grid power system. The system is evaluated according to its life cycle cost and green house gas emissions. While optimal sizing algorithms have been developed for different types of hybrid power systems, these often do not consider the socio-organisational drivers and limitations for inter-firm energy supply facilities. This work presents a list of recommendations for industrial park developers to determine and implement the park’s optimal power system. The current development of the project Eiland Zwijnaarde in Ghent provides the basis for a concrete case study in which the opportunities for an inter-firm power system are identified. The main results suggest that a significant share of renewable generation is part of the optimal configuration under varying mean electricity and gas prices. A medium sized cogeneration unit can compensate the renewables’ intermittent behaviour and lower the thermal energy cost. Large scale electrical storage is found not to be profitable under the used control structure and tariff scheme. A gradual ingress of firms in the park and the subsequent sloped annual energy demand has a negative effect on the fraction of shared facilities in the power system’s optimal configuration.

Keywords - On-grid hybrid power system, Optimal sizing, Eco-industrial Park, Industrial microgrid, Genetic Algorithm, Fuzzy logic Electrical design and operation of sustainable business parks Simon De Clercq Supervisor(s): Prof. dr. ir. Lieven Vandevelde, Dr. Brecht Zwaenepoel Ghent University, Faculty of Engineering and Architecture, Department of Electrical Energy, Metals, Mechanical constructions & Systems Academic year 2016-2017

Abstract— In this study a techno-economic optimisation is carried out ficantly reduce CO2 emissions and mitigate the effects of global on the design and operation of energy systems on newly-developed indus- warming. trial parks. An optimal sizing model is constructed that uses a genetic algorithm to find the optimal system configuration of an on-grid power An industrial park’s power system is characterised by loc- system. The system is evaluated according to its life cycle cost and green alised electric and thermal loads, high energetic consumption house gas emissions. While optimal sizing algorithms have been developed and spatial opportunities for the integration of renewable en- for different types of hybrid power systems, these often do not consider ergy sources. For these reasons, the prevailing paradigm for the socio-organisational drivers and limitations for inter-firm energy supply facilities. This work presents a list of recommendations for industrial electric distribution on an EIP is that of a microgrid. Accord- park developers to determine and implement the park’s optimal power sys- ing to [6], the microgrid concept assumes a cluster of loads tem. The current development of the project Eiland Zwijnaarde in Ghent and microsources operating as a single controllable system that provides the basis for a concrete case study in which the opportunities for an inter-firm power system are identified. The main results suggest that provides both power and heat to the local area. The compon- a significant share of renewable generation is part of the optimal config- ents in this energy cluster function independently in possible uration under varying mean electricity and gas prices. A medium sized interaction with the surrounding macrogrid [7]. This results in cogeneration unit can compensate the renewables’ intermittent behaviour enhanced local control, which can lower the cost of energy dis- and lower the thermal energy cost. Large scale electrical storage is found not to be profitable under the used control structure and tariff scheme. A tribution, aid the integration of renewable sources and thereby gradual ingress of firms in the park and the subsequent sloped annual en- reduce green house gas emissions [8]. Furthermore, the pos- ergy demand have a negative effect on the fraction of shared facilities in the sibility of islanding improves the system security of supply by power system’s optimal configuration. detaching the microgrid’s reliability from that of the macrogrid. Keywords—On-grid hybrid power system, Optimal sizing, Eco-industrial Park, Industrial microgrid, Genetic Algorithm, Fuzzy logic Several methods for the optimal sizing of hybrid power systems have been developed [9]. In many cases the power system is a stand-alone microgrid and consists of wind generation, I.INTRODUCTION solar PV, a micro source and electrical storage [10] [11] [12] Limitless economic growth, ecological collapse and resource [13] [14]. An industrial park’s high electrical demand, both in scarcity are forcing industry as a whole to rethink its funda- terms of quantity and quality, makes permanent islanding a com- mental principles and resort to more sustainable practices [1]. plicated design choice [15]. Furthermore, dense electrification As defined in the Brundtland Report [2], sustainable develop- in Flanders means that an industrial park can be connected to ment is development that meets the needs of the present without the macrogrid at a relatively low cost. An on-grid power system compromising the ability of future generations to meet their own is therefore considered in this study. needs. In this discourse, industrial symbiosis has emerged as While cogeneration offers high efficiency and flexibility [16], an approach in which traditionally separate industries collabor- only very few papers dealing with the optimal sizing of hybrid ate in order to find synergies that offer competitive advantages. power system take this feature into account [11]. A CHP model This can involve physical exchange of materials, energy, wa- is developed and included in the system model. ter, etc. [3]. Geographic proximity is an important facilitating A genetic algorithm is selected to carry out the optimisa- factor, which is why eco-industrial parks (EIP), with their col- tion process because of its computational efficiency in multi- lective infrastructure and spatial density of firms, have become objective optimisation problems with a high number of optim- an important topic of study in the field of sustainability [1]. ised variables [10] [11]. It has been used extensively in the op- The transition towards a carbon neutral industrial sector is timisation of hybrid power systems [9] [11]. driven by the global phenomenon of climate change. The IPCC II.HYBRIDPOWERSYSTEMMODEL warns that the continued emission of greenhouse gases will cause further warming of global climate patterns and result in The system model simulates the power flows in the microgrid long-lasting changes in all components of the climate system. on an hourly basis during its lifetime according to given load In Europe in 2014, 26% of CO2 equivalent emissions were due profiles, meteorological data and the system parameters. The to the consumption of electricity and gas [4]. Of all electricity system parameters consist of the installed solar power Pinst.solar, consumption in Europe, 36% is consumed by industry [5]. Op- the installed wind power Pinst.wind, the installed CHP power timizing industrial electricity consumption can therefore signi- Pinst.CHP, the battery storage capacity Einst.bat, and the maximum battery power Pinst.bat. The grid topology is shown in Figure 1. and the park’s electrical demand. If the difference is higher than half of the installed power Pinst.CHP, the unit is turned on and fulfils the demand as far as its operating region allows. This region is linearised and electric production is limited between 50% and 100% of the installed capacity [19] [20]. The nominal power-to-heat ratio is 1:1 [21] and there is always maximal heat production. The unit’s thermal and electrical efficiencies are variable and depend on the electrical set point, as shown in Table I. The heat demand that is not produced by the CHP installation is generated using an auxiliary gas boiler with a fixed efficiency of 0.80.

Table I: CHP prime mover efﬁciencies Electric load level ηel ηth ηtot 0.5 0.20 0.57 0.77 0.75 0.23 0.545 0.775 1 0.26 0.52 0.78

D. Battery storage Electrical power can be stored or withdrawn from an electrical storage unit. The storage is controlled using a fuzzy logic based control structure that was adapted from [22]. The main aim of the control structure is to charge the battery when the Fig. 1: Microgrid topology electricity price is low and discharge when the electricty price is high. This means that the battery can charge even though the power flow in the grid is negative, or discharge when the power A. Wind power flow is positive. However, under no circumstances should the Wind power can be modelled using wind speed data and a installed battery require fortification of the grid connection (and turbine’s power curve [10], [17] [18]. However, wind speed data thus a greater grid dependency). i 1 for the region of Flanders is hard to come by or simply does not The input parameters are the previous state of charge SOC − i exist. An alternative to wind speed data is historical data of wind and the instantaneous power flow at time i, PF . These are nor- power production [13] which in Belgium is easily accessible. malised and fuzzified according to 5 categories ranging between Data for on-shore wind generation in Belgium in 2015 is used very low and very high. The linguistic control rules are shown in to calculate the wind power production at every time step. The Table II. The resulting value Pfuzz is withdrawn from the battery normalised hourly time series of capacity factors is multiplied as far as the physical constraints allow. by the installed wind power Pinst.wind to obtain the amount of hourly wind production. Table II: Fuzzy control rules for battery storage i i i price PF Pfuzz i i P = capacity factor Pinst.wind (1) L / VP wind wind · M / Z The annual average capacity factor is equal to 0.244. H / VN The total generated electricity during one year is equal to the VH / VN sum of wind generation during every time step. (L) VH N ¬ (VH) VL P 8760 ¬ i (H) VL P Ewind = Pwind (2) ¬ VL: very low, L: low, M: medium, H: high, VH: very high i=1 X VN: very neg., N: neg, Z: zero, P: pos. VP: very pos. B. Solar power Historical solar production data for the region of East- Four constraints are considered. Flanders is used to calculate the hourly solar generation. The The power constraint dictates that power drawn from or fed annual average power factor is 0.119. to the battery during one time step can not exceed the maximum installed power Pinst.bat [23]. i i P = capacity factor Pinst.solar (3) The efficiency constraint takes into account the power loss solar solar · that occurs when energy is converted in the storage system. In C. Cogeneration theory the charging efficiency ηcharge and the discharging effi- The cogeneration unit follows the electric load according to ciency ηdischarge will have different values. However, separate the instantaneous difference between the renewable production values for these parameters are hard to measure [12] and battery manufacturers will therefore define a round trip efficiency for The initial investment cost depends on the size of the installed one entire charge and discharge cycle. In the model used here power and is equal to the sum of the investment cost for the sep- the charge and discharge efficiency are both assumed to be equal arate components according to the parameters in Table III. The to the square root of the round trip efficiency. ηbat = ηcharge = batteries have a double investment cost because of their smaller ηdischarge = √ηround = 0.90. life time. Storage systems exhibit self-discharge behaviour, which Cinitial = CAPEXsolar Pinst.solar + CAPEXwind Pinst.wind means that over time a certain fraction of the stored power will · · +2 CAPEX E + CAPEX P (9) be dissipated and lost [24] [17] [15]. The rate of self-discharge · c,bat · inst.bat p,bat · inst.bat per time step is denoted by the symbol σ. +CAPEX P CHP · inst.CHP By introducing a maximum depth of discharge of the elec- The annual cost is equal to the sum of the operational costs ac- trical storage unit, the longevity of the system can be prolonged cording to Table III. [24] [25]. Concretely this means that when the battery is at its minimum state of charge, it no longer allows power to be Cj = OPEX E + OPEX E annual solar · solar wind · wind withdrawn. SOCmax is in practice almost always equal to 1 and +OPEXf,CHP + OPEXv,CHP ECHP (10) SOCmin is here equal to 0.15. · The four constraints can mathematically be expressed as +grid cost + fuel cost shown below. i Electricity is bought (Pgrid < 0) at the instantaneous Belpex 0 < P exi < P (4) | bat| bat price plus an additional fee for the network costs. During time i i 1 i SOC = (1 σ) SOC − (5) steps with surplus electricity (Pgrid > 0) it is sold at the Belpex − · i price. The annual grid cost is calculated according to Equa- i 1 P ex i SOC − + η bat P ex < 0 tion 11. i bat Ebat bat SOC = · i (6) i 1 1 P ex i (SOC − + bat P ex > 0 8760 ηbat · Ebat bat i i i i (Pgrid < 0) (Belpex + fee) + (Pgrid > 0) Belpex (11) i · · SOC < SOC < SOC (7) i=1 min max X E. Demand Response The fuel consumption at every time step is calculated using the CHP set point, the CHP efficiency, and the auxiliary boiler’s In this study a heuristic approach is used to model demand consumption. The total annual fuel cost is the sum of the con- response. It looks at the daily interaction with the utility grid sumption multiplied by the gas price. and shifts a certain percentage of the total daily energy profile 8760 from times of high to times of low consumption. An example fuel consumptioni gas price (12) for the original and resulting power flow for one week is shown · i=1 in Figure 2. Using this method, a shift of 2.5% of the daily X energy interaction is found to correspond to an annual power peak reduction by 10.7%. Table III: Economic parameters [27] [28] Battery Solar Wind CHP Capacity Power 500 without DR CAPEX (e/kW) 1570 1911 3940 175 175 with DR 0 OPEXv (e/kWh) 0.021 0.017 0.0108 - - OPEXf (e/year) - - 9345 - - -500 Lifetime (year) 25 25 25 12.5

-1000

Grid power in kW -1500 B. Emissions evaluation

-2000 The microgrid’s annual CO2 emissions are calculated according to the total annual production per source and their -2500 0 20 40 60 80 100 120 140 160 technology-speciﬁc reference values (Table IV). hour tot rel.solar rel.wind Fig. 2: Demand response model CO2 = CO2 Esolar + CO2 Ewind+ · · (13) rel.CHP rel.grid CO2 ECHP + CO2 Egrid III.SYSTEM EVALUATION AND OPTIMISATION · · In the case of a netto export of electricity to the grid, the emis- A. Economic evaluation sions due to the additional electricity production are subtracted The system is evaluated according to its life cycle cost (LCC). from the microgrid’s emissions. The relative share of each of This consists of the initial investment cost Cinitial and the dis- the microgrid’s generation units allows a relative value for the j counted sum of the annual cost Cannual for each year [10] [26] CO2 emissions per kWh of produced power to be calculated. [15]. rel.solar rel.wind rel.CHP LT j CO E + CO E + CO E C COmix = 2 · solar 2 · wind 2 · CHP LCC = C + annual (8) 2 E + E + E initial (1 + r)j solar wind CHP j=1 (14) X rel.grid For netto export, this value replace the value of CO2 in Figure 5 shows the microgrid’s hourly production proﬁles for Equation 13. one week in the summer. There are strong daily generation peaks due to the solar power production. These are almost dir- Table IV: Relative CO2 emissions [29] ectly translated into energy export peaks towards the grid. The Solar Wind CHP Grid CHP mainly works at night to compensate the absence of solar rel CO2 (gCO2/kWh) 41 11 300 210 generation. There is a very clear netto positive power exchange with the grid. The value for the LCC in this scenario is 188 million e. IV. EILAND ZWIJNAARDE 100.24 gCO2 is emitted per consumed kWh. The LCC for a Eiland Zwijnaarde is an industrial park in the region of Ghent reference case with the same electricity and gas buying price e that is currently under development. The developers are assess- but without on-site production is 243 million . The reference ing the possibilities of collective energy infrastructure and it is emissions are equal to the grids emissions, 210 gCO2/kWh. The therefore an appropriate case study for the developed model. proposed power system thus offers an improvement according The site’s annual electric consumption is estimated around 60 to both evaluation methods.

GWh. and the annual thermal demand at 40 GWh. The electric ×104 load profile is based on the consumption profile of a technology 1.5 park in the region of Ghent (Figure 3). The thermal load profile is a generic load profile from the Flemish Energy Regulator 1 (Figure 4). The peak electrical demand is 12 MW and the peak 0.5 thermal demand is 14 MW.

12000 15000 Average power (kW) 11000 -0.5 P solar 10000 P wind 10000 9000 -1 P CHP P 8000 grid Load profile 7000 -1.5 5000 20 40 60 80 100 120 140 160 Electric load profile (kW) 6000 Thermal load profile (kW) Time (hours) 5000 Fig. 5: Hourly production profile over one week 4000 0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 hour hour Fig. 3: Electric load profile Fig. 4: Thermal load profile A. DR Demand response is included in the model as described The upper boundary for the optimal size of the installed tech- above. On a daily basis 2.5% of the total electrical demand is nologies are derived from the physical boundaries of the site. In shifted from times of high demand to times of low demand. total there is 350000 hectares available for companies. At a con- The resulting optimal installed power for each technology is 2 shown in Figure 6. There is a general decrease in installed power servative estimate of 10m /kWinst. for solar power, this means that the maximum amount of solar peak power that can be in- compared to the optimisation results without demand response. stalled is 35MW. For wind power it is estimated that there is The total size reduction is about 15%. The reduction is most space to install 3 large turbines. At 3.3 MW/turbine, the res- pronounced for solar power and the grid connection; both de- ulting upper boundary for installed peak power is 10 MW. The crease by more than 3MW. In terms of energy, shown in Figure upper boundary for CHP is the electric peak demand, 10 MW. 7, there is less solar power production and a subsequently smal- There is no upper boundary for the installed battery power or its ler energy export to the utility grid. Overall, there is a smaller energetic capacity. interaction with the utility grid while maintaining a high share of renewable power. V. CASE STUDY AND SIMULATION RESULTS The optimal sizing results for an average electricity buying 30 DR No DR price of 115 e/MWh and a gas price of 63 e/MWh are tabulated 20 below. Initially, demand response is not considered.

Psolar : 24.70 MW Esolar : 25700 MWh/year (MW) 10 Pwind : 9.91 MW Ewind : 21150 MWh/year Installed power 0 PCHP : 3.56 MW ECHP : 17070 MWh/year Solar Wind CHP Grid Pgrid : 17.40 MW Egrid : -3925 MWh/year Fig. 6: Installed power with demand response Ebat : 18 kWh There is a high share of both solar and wind power in the total annual electricity generation. The amount of wind power is just below the site’s upper boundary and there is a netto export of B. Gradual ingress electricity to the utility grid. The amount of installed battery In the previous sections the hybrid power system was sized power is negligible on this scale. according to an equal energetic demand for every year of its life- 104 104 · · 3 DR No DR 3 Ingress No ingress 2 2 1 1

(MWh/year) 0 (MWh/year) 0 Generated power Generated power -1 -1 Solar Wind CHP Grid Solar Wind CHP Grid Fig. 7: Generated power with demand reponse Fig. 9: Generated power with gradual ingress time. However, a gradual ingress of companies on the site will rapidly for an increasing value of LCC. On average, a reduc- result in a much smaller load profile during the first years of the tion of 1 gCO2/kWh comes at an additional cost of 117000 e park. In the model, the first year’s electric and thermal demand over the lifetime of the installation. This corresponds to 4680 e are set to 10% of the nominal value. During the following ten on an annual basis. For an LCC greater than 19 million e the years the demands increase by 10% every year. After 10 years curve is practically linear and the total cost of saving an ad- the electric and thermal demand are at their nominal value. The ditional gC02/kWh is equal to 340000 e, or an annual cost of power system is sized once and the installed capacities are fixed 13600 e/year . during the entire lifetime of the installation.

The simulation results are shown in Figures 8 and 9. There is Pareto front a clear absence of cogeneration in the optimised system. In- 100 stead, all of the thermal energy is provided by the auxiliary 95 boiler. The CHP unit is controlled according to the electrical 90 demand, which means that during the ﬁrst years it will prac- 85 tically never be turned on. Instead the electrical power mainly 80 comes from the grid and the wind turbines. While in the previ- 75 ous optimisations there was always a netto export of electricity 70 to the grid, in this case there is a large netto import. This grid 65 dependency does not result in a larger grid connection; the grid’s GHG emissions in gCO2/kWh 60 installed power decreases from around 18 MW to 13 MW. This 55 is due to the lower installed solar fraction. Solar power gives 50 1.88 1.9 1.92 1.94 1.96 1.98 2 2.02 2.04 large production peaks during summer which requires a strong Stop Pause ×108 grid connection to inject the energy surplus on the grid. De- Fig. 10: Pareto curve creasing the solar fraction thus means that the grid connection can be made smaller. Figure 11 shows each technology’s annually generated elec- The relative CO2 emissions are 95.7 gCO2/kWh. This is tricity for every point on the pareto front. For low LCC, the slightly lower than the reference scenario without gradual in- reduction in CO2 is due to a decrease in cogeneration and an gress. increase in solar power. This corresponds to the steepest part of the pareto curve, which means that this reduction comes at 30 the lowest cost per gCO2 saved. The wind power fraction and Ingress No ingress grid export stay relatively constant. For increasing LCC, the 20 generated solar power reaches its maximum. For a further CO2

(MW) 10 reduction the installed power of the CHP keeps decreasing and the energy deficit is compensated by less electricity exports to Installed power 0 the grid. This way the emissions can be reduced down to about Solar Wind CHP Grid 54 gCO2/kWh. In comparison, the average in electricity in Bel- Fig. 8: Installed power with gradual ingress gium are between 180 and 210 gCO2/kWh [30] and in Europe this is around 320 gCO2/kWh [31]. For very low values of CO2 the netto power exchange with the grid becomes positive. C. Multi-dimensional optimisation VI.CONCLUSION A multi-objective optimisation is carried out that finds the nondominated solution set for the LCC and the CO2 emissions The optimal system is assessed for a range of scenarios. The per kWh. The pareto curve is shown in Figure 10. results suggest a large scale roll-out of renewable generation in The configuration with the lowest LCC and the highest CO2 practically all cases. For maximal installed renewable produc- lies around 18.8 million e. This corresponds to the result of the tion the renewable power sources are theoretically able to supply single-objective optimisation. The CO2 emissions initially drop all the park’s annual energy demand. However, the intermittent ×107 3.5 [9] S. R. Tito, T. T. Lie, and T. N. Anderson, “Optimal sizing of a wind- photovoltaic-battery hybrid renewable energy system considering socio- 3 demographic factors,” Solar Energy, vol. 136, pp. 525–532, 2016. [10] E. Koutroulis, D. Kolokotsa, A. Potirakis, and K. Kalaitzakis, “Method- 2.5 ology for optimal sizing of stand-alone photovoltaic/wind-generator sys- 2 tems using genetic algorithms,” Solar Energy, vol. 80, no. 9, pp. 1072– E solar 1088, 2006. 1.5 E wind [11] M. S. Ismail, M. Moghavvemi, and T. M. I. Mahlia, “Genetic algorithm E CHP based optimization on modeling and design of hybrid renewable energy 1 E grid systems,” Energy Conversion and Management, vol. 85, pp. 120–130, 0.5 2014. Generated power (kWh) [12] H. Yang, Z. Wei, and L. Chengzhi, “Optimal design and techno-economic 0 analysis of a hybrid solar-wind power generation system,” Applied Energy, -0.5 vol. 86, no. 2, pp. 163–169, 2009. [13] B. Zhao, X. Zhang, P. Li, K. Wang, M. Xue, and C. Wang, “Optimal sizing, -1 operating strategy and operational experience of a stand-alone microgrid 95.6gCO /kWh 71.74gCO /kWh 53.94gCO /kWh 2 2 2 on Dongfushan Island,” Applied Energy, vol. 113, pp. 1656–1666, 2014. [14] R. Dufo-Lopez´ and J. L. Bernal-Agust´ın, “Multi-objective design of Fig. 11: Generated power on pareto front PV-wind-diesel-hydrogen-battery systems,” Renewable Energy, vol. 33, no. 12, pp. 2559–2572, 2008. [15] W. L. Theo, J. S. Lim, S. R. Wan Alwi, N. E. Mohammad Rozali, W. S. Ho, and Z. Abdul-Manan, “An MILP model for cost-optimal planning of behaviour of these sources and the current price structure means an on-grid hybrid power system for an eco-industrial park,” Energy, vol. that a solid grid connection is in generally favourable. 116, pp. 1423–1441, 2016. [16] B. Zwaenepoel, J. Vansteenbrugge, T. L. Vandoorn, G. Van Eetvelde, and A small amount of demand response can have a large impact L. Vandevelde, “Renewable energy balancing with thermal grid support,” on the size of the installation. A power system with DR is more Chemical Engineering Transactions, vol. 35, pp. 535–540, 2013. self-sufficient without increasing the fraction of non-renewable [17] Y. Li, X. Chen, J. Xiao, and X. Wei, “Optimal configuration for distributed generations in micro-grid system considering diesel as the main control production. source,” in International Conference on Advanced Mechatronic Systems, A gradual ingress of firms in the park and the subsequent 2014, pp. 552–556. [18] S. M. M. Tafreshi, H. A. Zamani, S. M. Ezzati, M. Baghdadi, and H. Va- sloped annual energy demand has a very significant effect on hedi, “Optimal unit sizing of distributed energy resources in microgrid the power system’s optimal configuration. Dynamically sizing using genetic algorithm,” Proceedings - 2010 18th Iranian Conference on the site’s power system according to a changing annual energy Electrical Engineering, ICEE 2010, pp. 836–841, 2010. [19] H. Wang, W. Yin, E. Abdollahi, R. Lahdelma, and W. Jiao, “Modelling and demand could be a future topic of research to address this issue. optimization of CHP based district heating system with renewable energy The implemented storage control structure is not able to offer production and energy storage,” Applied Energy, vol. 159, pp. 401–421, 2015. any economic incentives for the large scale deployment of bat- [20] M. Nazari-Heris, S. Abapour, and B. Mohammadi-Ivatloo, “Optimal eco- tery storage. This suggests that new remuneration methods are nomic dispatch of FC-CHP based heat and power micro-grids,” Applied needed in order to successfully introduce this technology. Thermal Engineering, vol. 114, pp. 756–769, 2017. [21] P. Ghadimi, S. Kara, and B. Kornfeld, “The optimal selection of on-site A multi-objective optimisation identifies the specific cost of CHP systems through integrated sizing and operational strategy,” Applied decreasing CO2 emissions. Increasing the solar share and redu- Energy, vol. 126, pp. 38–46, 2014. [22] J. P. Fossati, A. Galarza, A. Martin-Villate, J. M. Echeverria, and cing the size of the CHP installation results in reduced emissions L. Fontan, “Optimal scheduling of a microgrid with a fuzzy logic con- but comes at an additional cost. trolled storage system,” International Journal of Electrical Power and En- While in all scenarios there is a high share of renewable gen- ergy Systems, vol. 68, pp. 61–70, 2015. [23] J. P. Fossati, A. Galarza, A. Martin-Villate, and L. Fontan, “A method for eration, there is limit to the physical amount of energy that can optimal sizing energy storage systems for microgrids,” Renewable Energy, be generated on-site. Solar and wind power clearly play an im- vol. 77, pp. 539–549, 2015. portant role in the optimal power system, but their share is lim- [24] H. Yang, W. Zhou, L. Lu, and Z. Fang, “Optimal sizing method for stand- alone hybrid solar-wind system with LPSP technology by using genetic ited by the site’s physical boundaries. Demand reduction should algorithm,” Solar Energy, vol. 82, no. 4, pp. 354–367, 2008. therefore be considered in order to make the park independent [25] A. Chaouachi, R. M. Kamel, R. Andoulsi, and K. Nagasaka, “Multiobject- ive Intelligent Energy Management for a Microgrid,” IEEE Transactions of exhaustive energy sources. on Industrial Electronics, vol. 60, no. 4, pp. 1688–1699, 2013. [26] A. Gonzalez,´ J. R. Riba, A. Rius, and R. Puig, “Optimal sizing of a hy- REFERENCES brid grid-connected photovoltaic and wind power system,” Applied En- ergy, vol. 154, pp. 752–762, 2015. [1] H. Zhao, H. Zhao, and S. Guo, “Evaluating the comprehensive benefit [27] International Energy Agency, Nuclear Energy Agency, and Organisation of eco-industrial parks by employing multi-criteria decision making ap- for Economic Co-operation and Development, “Projected Costs of Gener- proach for circular economy,” Journal of Cleaner Production, vol. 142, ating Electricity 2015,” Tech. Rep., 2015. pp. 2262–2276, 2017. [28] K. Darrow, R. Tidball, J. Wang, and A. Hampson, “Catolog of CHP [2] United Nations World Commission on Environment and Development, Technologies,” U.S. Environmental Protection Agency, Tech. Rep. March, “Our Common Future,” 1987. 2015. [3] M. R. Chertow, “Industrial Symbiosis: Literature and Taxonomy,” Annual [29] IPCC, “Climate Change 2013 - The Physical Science Basis,” Tech. Rep., Review of Energy and the Environment, vol. 25, no. 1, pp. 313–337, 2000. 2014. [4] Eurostat, “Energy balance sheets,” pp. 345–52, 2016. [30] M. Messagie, J. Mertens, L. Oliveira, S. Rangaraju, J. Sanfelix, T. Coose- [5] European Environmental Agency, “Final electricity consumption by sector mans, J. Van Mierlo, and C. Macharis, “The hourly life cycle carbon foot- , EU-27,” p. European Environment Agency, 2013. print of electricity generation in Belgium, bringing a temporal resolution [6] R. Lasseter, “MicroGrids,” in IEEE Power Engineering Society Winter in life cycle assessment,” Applied Energy, vol. 134, no. August, pp. 469– Meeting. Conference Proceedings, vol. 1, 2002, pp. 305–308. 476, 2014. [7] K. V. Hende and C. Wouters, “The Regulation of Microgrids in Liberalized [31] F. Millo, L. Rolando, R. Fuso, and F. Mallamo, “Real CO2 emissions be- Electricity Markets in the EU and East Asia,” European Networks Law and nefits and end user’s operating costs of a plug-in Hybrid Electric Vehicle,” Regulation Quarterly, vol. 3, pp. 190–204, 2014. Applied Energy, vol. 114, pp. 563–571, 2014. [8] R. Hanna, M. Ghonima, J. Kleissl, G. Tynan, and D. G. Victor, “Evaluating business models for microgrids: Interactions of technology and policy,” Energy Policy, vol. 103, no. August 2016, pp. 47–61, 2017. Contents

1 Introduction 1

1.1 Eco-industrial parks ...... 2

1.1.1 Deﬁnitions and concepts ...... 2

1.1.2 Drivers and limitations of eco-industrial parks ...... 3

1.1.3 Inter-ﬁrm energy supply ...... 5

1.1.4 Industrial microgrids ...... 6

1.2 Eiland Zwijnaarde ...... 9

1.2.1 Eiland Zwijnaarde as an eco-industrial park ...... 10

1.2.2 Vision of project ...... 10

1.2.3 Current state of aﬀairs ...... 13

2 Microgrids in the Belgian electricity sector 15

2.1 The liberalised electricity market ...... 15

2.2 Price of electricity ...... 18

2.2.1 Commodity price ...... 19

2.2.2 Transmission tarrifs ...... 20

2.2.3 Distribution tarrifs ...... 21

2.2.4 Industrial energy contract ...... 21

2.3 Energy policy in Flanders ...... 22

2.4 CO2 market ...... 24

2.5 Grid structure and microgrid opportunities ...... 25

2.5.1 Functionality and economic incentives ...... 25

xiii 2.5.2 Control ...... 27

2.5.3 Ownership ...... 27

3 Hybrid Power systems 30

3.1 Wind power ...... 31

3.1.1 Wind turbine power curve ...... 32

3.1.2 Wind speed data ...... 33

3.1.3 Dynamic economic analysis ...... 35

3.2 Solar power ...... 37

3.2.1 PV-curve & Maximum power point (MPP) ...... 38

3.2.2 Dynamic economic analysis ...... 40

3.3 Cogeneration ...... 42

3.3.1 Cogeneration technologies ...... 43

3.3.2 Modelling and sizing ...... 45

3.3.3 Modes of operation ...... 50

3.3.4 Dynamic techno-economic analysis ...... 51

3.3.5 System evaluation ...... 53

3.3.6 Reference case ...... 54

3.3.7 Electric load following: ...... 54

3.3.8 Thermal load following: ...... 56

3.3.9 Conclusion on CHP model ...... 58

3.4 Electrical storage ...... 58

3.4.1 Modelling ...... 59

3.4.2 Battery control ...... 62 3.5 Demand-side management ...... 68

3.5.1 Demand response modeling ...... 69

4 Optimal design and sizing 71

4.1 Evaluation of electricity production system ...... 71

4.1.1 Economic evaluation ...... 71

4.1.2 Technical evaluation and constraints ...... 75

4.2 Optimised variables ...... 77

4.3 Multi-objective optimisation ...... 78

4.3.1 Pareto front ...... 78

4.4 Genetic Algorithm ...... 79

5 Simulation results and discussion 82

5.1 Energy key ﬁgures and model parameters ...... 82

5.2 Sensitivity analysis ...... 83

5.2.1 System evaluation ...... 84

5.2.2 Installed power per technology ...... 85

5.2.3 Average weekly power ﬂow ...... 87

5.3 Future prices ...... 90

5.4 Impact of ETS market ...... 93

5.5 Impact of demand response ...... 94

5.6 Impact of gradual ingress ...... 96

5.7 Multi-dimensional analysis ...... 100

6 Conclusion 103

List of Abbreviations and Symbols

ARP Access responsible party ACS Annualized cost of system BREAAM Building Research Establishment Environmental Assessment Method

Cinitial Initial investment cost

Cannual Annual system cost

CAPEXsolar Capital expenditure for solar power

CAPEXwind Capital expenditure for wind power

CAPEXCHP Capital expenditure for CHP unit

CAPEXc,bat Capital expenditure for battery capacity

CAPEXp,bat Capital expenditure for battery power CHP Combined heat and power

CO2,mix Average CO2 emissions per kWh of electricity produced locally in the microgrid COE Cost of electricity DR Demand response DSM Demand side management DSO Distribution system operator

ECHP Annually produced electrical CHP power

Egrid Annual power exchange with macrogrid

Esolar Annually produced solar power

Ewind Annually produced wind power

Einst.bat Maximum amount of energy that can be stored in the storage system EE Embodied energy EES Electrical energy storage EIP Eco-industrial park ESCO Energy service company ETS Emission trading system GHG Green house gas LCC Life cycle cost i LPE Instantaneous electrical demand i LPT Instantaneous thermal demand LPSP Loss of power supply probability LT Lifetime MPP PV maximum power point MOC Mean operation cost

OPEXf,CHP Fixed operational expenditure CHP

OPEXsolar Operational expenditure solar power

OPEXv,CHP Variable operational expenditure CHP

OPEXwind Operational expenditure wind power OTC Over the counter PB Power balance i P exbat Instantaneous power exchange with storage system PF i Instantaneous diﬀerence between the electric production and the electric consumption i PFnorm Normalised instantaneous power ﬂow Pgrid1 Grid interaction because of the battery’s power limitation

Pgrid2 Grid interaction because of the battery’s state of charge limitation

PI/O Reference value for CHP on-oﬀ decision i PCHP Instantaneous CHP electrical power production i Pfuzz Battery’s instantaneous economic set point i Pgrid Instantaneous interaction with macrogrid i Psolar Instantaneous solar power production i Pwind Instantaneous wind power production Pinst.bat Maximum amount of power that can be drawn or stored in the battery during one time step

Pinst.CHP Installed electrical CHP power

Pinst.grid Installed grid power

Pinst.solar Installed solar power

Pinst.wind Installed wind power PCC Point of common coupling P-Q ratio Power to heat ratio i pricenorm Normalised instantaneous electricity price PV Photovoltaic i QCHP Instantaneous CHP heat production i Qboiler Instantaneous boiler heat production r Interest rate REP Renewable energy penetration ROI Return on investment SOCi State of charge

SOCmin EES minimum state of charge

SOCmax EES maximum state of charge SPV Special purpose vehicle TSO Transmission system operator UPS Uninterruptible power supply vi Instantaneous wind speed vci Cut-in wind speed vr Rated wind speed vco Cut-out wind speed σ EES hourly self-discharge coeﬃcient

ηbat EES charging and discharging eﬃciency

ηel Electrical CHP eﬃciency

ηth Thermal CHP eﬃciency

ηtot Total CHP eﬃciency List of Figures

1.1 Classiﬁcation of industrial parks ...... 3

1.2 Microgrid topology ...... 8

1.3 Map of the site Eiland Zwijnaarde ...... 9

1.4 Development scenarios for Eiland Zwijnaarde ...... 11

1.5 Trias Energetica ...... 13

2.1 Lay-out of liberalised electricity market ...... 16

2.2 Breakdown of the price of electricity ...... 19

2.3 Hourly Belpex price in 2015 ...... 19

2.4 Projection of ETS CO2 price ...... 25

3.1 Wind turbine power curve ...... 32

3.2 Wind speed proﬁles based on the log and power law ...... 34

3.3 Historical wind production data ...... 35

3.4 Hourly load proﬁle ...... 37

3.5 Hourly wind power production ...... 37

3.6 Hourly power exchange with the grid ...... 37

3.7 Power ﬂows over one week ...... 37

3.8 3D Simulation results for wind power ...... 38

3.9 Simulation results for speciﬁc grid price ...... 39

3.10 PV power-voltage characteristic ...... 40

3.11 Historical solar production data ...... 40

3.12 Hourly solar power production ...... 41

xix 3.13 Power ﬂows over one week ...... 41

3.14 3D Simulation results solar power ...... 42

3.15 Simulation results for speciﬁc grid price ...... 42

3.16 Lay-out of a CHP system ...... 43

3.17 CHP operating region ...... 48

3.18 Operational cost of CHP ...... 49

3.19 Linearised CHP operating region ...... 50

3.20 Hourly electric load proﬁle ...... 52

3.21 Hourly thermal load proﬁle ...... 52

3.22 Hourly Belpex price ...... 52

3.23 Electric load following control algorithm ...... 55

3.24 Simulation results with electric load following ...... 56

3.25 Thermal load following control algorithm ...... 57

3.26 Simulation results with thermal load following ...... 57

3.27 Battery control algorithm - maximal usage ...... 64

3.28 Battery control algorithm - fuzzy logic ...... 65

3.29 Membership functions for power ﬂow ...... 67

3.30 Membership functions for electricity price ...... 67

3.31 Membership functions for Pfuzz ...... 68

3.32 Demand response over one day ...... 70

3.33 Demand response over one week ...... 70

3.34 Pgrid without DR (entire year) ...... 70

3.35 Pgrid with DR (entire year) ...... 70 4.1 Graphical representation of a pareto front ...... 79

4.2 Flowchart of a genetic algorithm ...... 80

5.1 Hourly electric load proﬁle ...... 83

5.2 Hourly thermal load proﬁle ...... 83

5.3 Life cycle cost ...... 85

5.4 Cost of electricity ...... 85

5.5 CO2 emissions ...... 86

5.6 Power exchange with grid ...... 86

5.7 Installed solar power ...... 87

5.8 Installed wind power ...... 87

5.9 Installed CHP power ...... 87

5.10 Installed storage capacity ...... 87

5.11 Weekly power ﬂow during the entire year ...... 88

5.12 Hourly simulation during one week in winter ...... 89

5.13 Hourly simulation during one week in summer ...... 90

5.14 Life cycle cost ...... 91

5.15 Cost of electricity ...... 91

5.16 CO2 emissions ...... 91

5.17 Installed solar power ...... 92

5.18 Installed wind power ...... 92

5.19 Installed CHP power ...... 92

5.20 Installed batter capacity ...... 92

5.21 Installed power with ETS ...... 94 5.22 Generated power with ETS ...... 94

5.23 Installed power with DR ...... 96

5.24 Generated power with DR ...... 96

5.25 Eﬀect of ingress on energetic demand ...... 97

5.26 Installed power with ingress ...... 98

5.27 Generated power with ingress ...... 98

5.28 Weekly power ﬂow for the fourth and tenth year ...... 99

5.29 Hourly power curves over one week in summer for diﬀerent years ...... 100

5.30 2-dimensional pareto front ...... 101

5.31 Installed power on pareto front ...... 101

5.32 Generated power on pareto front ...... 101 List of Tables

2.1 Example of an industrial energy contract ...... 22

3.1 Overview of technologies in studied literature ...... 31

3.2 Commercial wind turbines ...... 33

3.3 Simulation parameters for wind power ...... 36

3.4 Overview simulation parameters for solar power ...... 41

3.5 Overview of CHP technologies ...... 44

3.6 CHP cost function coeﬃcients ...... 49

3.7 Cogeneration prime mover eﬃciencies ...... 50

3.8 Overview simulation parameters for cogeneration case ...... 53

3.9 Technical parameters for diﬀerent storage technologies ...... 59

3.10 Economic parameters for storage technologies ...... 60

3.11 Fuzzy control rules for battery storage ...... 68

3.12 Demand response simulation results ...... 70

4.1 Overview of optimisation methods in studied literature ...... 72

4.2 List of abbreviations and occurrence ...... 73

4.3 CO2 emissions per technology in gCO2eq/kWh ...... 76

5.1 Predicted energetic consumption for diﬀerent scenarios ...... 82

5.2 Economic simulation parameters ...... 84

5.3 Future economic simulation parameters ...... 90

5.4 Simulation results with ETS ...... 93

5.5 Simulation results with DR ...... 95

xxiii xxiv

5.6 Simulation results with gradual ingress ...... 97

1 Sensitivity analysis for current market prices ...... 119

2 Sensitivity analysis for future market prices ...... 120 Chapter 1

Introduction

Limitless economic growth, ecological collapse and resource scarcity are forcing industry as a whole to rethink its fundamental principles and resort to more sustainable practices [1]. As defined in the Brundtland Report [2], sustainable development is development that meets the needs of the present without compromising the ability of future generations to meet their own needs. In this discourse, industrial symbiosis has emerged as an approach in which traditionally separate industries collaborate in order to find synergies that offer competitive advantages. This can involve physical exchange of materials, energy, water, etc. [3]. Geo- graphic proximity is an important facilitating factor, which is why eco-industrial parks (EIP), with their collective infrastructure and spatial density of firms, have become an important topic of study in the field of sustainability [1].

The transition towards a carbon neutral industrial sector is driven by the global phenomenon of climate change. The IPCC warns that the continued emission of greenhouse gases will cause further warming of global climate patterns and result in long-lasting changes in all components of the climate system. According to their 2014 assessment report, these change will increase the likelihood of severe, pervasive and irreversible impacts for people and ecosystems [4]. In

Europe in 2014, 26% of CO2 equivalent emissions were due to the consumption of electricity and gas [5]. Of all electricity consumption in Europe, 36% is consumed by industry [6].

Optimizing industrial electricity consumption can therefore signiﬁcantly reduce CO2 emissions and mitigate the eﬀects of global warming.

The aim of this study is to carry out a techno-economic analysis on the design and operation of energy systems on newly-developed industrial parks. A microgrid model is constructed that optimises the system configuration according to its life cycle cost and green house gas emissions. The socio-organisational framework is assessed by means of a literature study which results in a list of recommendations for park developers. The current development of the project Eiland Zwijnaarde in Ghent provides the basis for a concrete case study in which the opportunities for an inter-firm power system are identified.

1 Chapter 1. Introduction 2

1.1 Eco-industrial parks

1.1.1 Deﬁnitions and concepts

What sets EIPs apart from conventional business parks is that the ecological footprint and carbon dioxide emissions of the individual businesses as well as the park as a whole are actively reduced. Energy and construction technologies are employed to optimise energy eﬃciency and material cycles. An integrated management system facilitates inter-ﬁrm exchange of energy, materials and information. A regulatory system encourages the companies to meet their performance goals [7].

Four main definitions in the field of eco-industrial parks can be identified. They are schem- atically represented in Figure 1.1.

Sustainable industrial parks A sustainable industrial park is the most far-reaching concept in terms of sustainability and collectivity in the context of industrial park design. Technological, economic, social and spatial opportunities oﬀered by the geographical proximity are combined in ways that go beyond the ﬁrms’ individuality [8]. This results in a collective and integrated approach regarding the supply and exchange of energy and materials, use of equipment and facilities, etc. Sustainable industrial parks are well-integrated in the surrounding area and require extensive measures regarding ownership and stakeholders relationships.

Eco-industrial parks In EIPs, individual companies exploit inter-firm synergies in term of materials, energy and water. There is a strong focus on industrial symbiosis between the participants, although the firms pursue their individual objectives. An eco-industrial park can be seen as the composition of a number of industrial symbiosis projects between different firms in the same location [9]. It forms an industrial ecosystem in which the consumption of energy and materials is optimised, waste generations is minimised and residues, whether in terms of materials or energy, are optimally reinjected in the value chain [10].

Green industrial parks A green industrial park is a collection of ﬁrms that individually strive towards sustainability. Possibilities to exploit inter-ﬁrm synergies are not actively investigated. The majority of sites in Flanders with an eco-label belong to this group [11].

Low carbon industrial parks Low carbon industrial parks is the combination of green industrial parks and eco- industrial parks. Firms focus on renewable energy and clean processes on an individual Chapter 1. Introduction 3

level while looking for opportunities of interaction with neighbouring sites.

Sustainable Eco-industrial Green industrial parks parks industrial parks

Low carbon industrial parks

Figure 1.1: Classiﬁcation of industrial parks [11]

1.1.2 Drivers and limitations of eco-industrial parks

Since the concept of industrial symbiosis was ﬁrst used in academic literature in the 1980’s, there have been many attempts to harness its advantages [12]. A review of literature is carried out to identify the drivers and limitations of the implementation of industrial ecology concepts in industrial parks. Many of the cases cover a broad range of technologies, from energy and resource supply to transport, employee facilities and water treatment. The scope of this study is conﬁned to the possibilities regarding collective supply of electrical power. The fact that eco-industrial parks are such multi-disciplinary entities makes them hard to study. The technical aspects, the business framework around them and the social interaction with the park’s surroundings form an extensive socio-organisational web [7].

A list of drivers and opportunities of EIPs is presented, as adapted from [13], [12], [7] and [14].

While the ideological basis for the development of an EIP can be environmental, the most important driver for a successful project is the financial gain. Business performance and environmental performance should therefore both play an important role in the design of the park and inter-firm cooperation will only then offer advantages compared to regular (more wasteful) business park design.

Two ways of thinking exist as to whether EIPs should be artificially engineered or self- organised. The first perspective focuses on the technical feasibility of the inter-firm linkages, while the second believes that organic growth of connections will result in Chapter 1. Introduction 4

enhanced ownership and a more resilient system. What is most important is that the system is fully integrated in its surrounding and that it possesses suﬃcient ﬂexibility to survive in a dynamic economic environment.

Ideally, the initiative for inter-firm cooperation should come from the firms themselves and not from any other authority. The role of the park management should be to facilitate information flow between the firms and to provide a platform for open communication.

Physical proximity and complementary proﬁles in terms of energy and materials is important and direct economic competition between ﬁrms should be avoided [15]. In any case, a level of trust between the participants is crucial.

An active participation of the firms in collective projects ensures continuity and sus- tained ownership. Participation from additional stakeholders such as other companies, the local community, environmental organisations, as well as technical experts from different fields can be a strong asset in the success of a project.

During the planning phase of the site, the focus should be primarily on utility sharing. The subsequent economic and environmental advantages will convince the companies of the beneﬁts of symbiotic exchanges [12]. In a second instance, energy, water and material waste exchanges can be considered. Finally more company speciﬁc and economically challenging projects can be developed.

The most famous example of a successful EIP is the Kalundborg Park in Denmark. It is the first realisation of industrial symbiosis that was published in literature [3]. Nine different firms, including a coal-fired power station, an oil refinery and a pharmaceutical plant are connected through a network of eleven physical linkages. Remarkable is that the inter-connections were not conceived from the design of the park but instead grew gradually through independent and economically driven actions [16]. The Asnaes Power Station has a very central role in the system, which shows the fundamental role of energy supply in EIP design. The plant provides electricity as well as heat and process steam to its industrial and residential neighbours. The power station is still coal fired but a transition to biomass is planned in the near future [17].

While there are inherent advantages for interactions between ﬁrms, technical, economic, in- formational, organisational and regulatory barriers can be identiﬁed [12].

The main limitation of EIP parks is the potential for system fragility. If one of the park’s main firms decides to re-locate, other companies might need to find other sources for their raw materials thus affecting the functioning of the entire chain. Electrical power however, as opposed to physical primary resources, is homogeneous independently of Chapter 1. Introduction 5

its source, and therefore collective power supply is less aﬀected by this fragility. Dis- tribution systems though will be sized and reinforced according to the type of installed generation, which means that a change in the production might result in poor power quality and reliability. This is worse in the case of heat exchange, because quality and temperature are very important parameters for process heat. One way of mitigating this vulnerability is a diversiﬁcation of sources [13].

Another limitation is that the price of inputs and output, such as electricity and other energy forms, vary over time. This means that the proﬁtability of inter-ﬁrm projects is subject to uncertainty. Furthermore, low market prices for (non-renewable) energy sources means that companies often do not have enough incentives to invest in collective renewable energy projects.

In the energy sector the legal framework is often a hindrance. Administrative barriers, public opposition and ever-changing or non-existent legislation create an unfavourable environment for the development of inter-ﬁrm energy supply. This limitation is more broadly assessed in chapter 2 on microgrids in the Belgian market.

1.1.3 Inter-ﬁrm energy supply

The high investment cost and significant green house gas emissions in industrial energy supply mean that sharing facilities between firms is a promising approach [18]. The collective nature of electrical distribution means that several advantages of an inter-firm power grid can be identified [18].

Liberalisation of energy markets might open up new opportunities for inter-ﬁrm energy projects. Using an existing installation to participate in new market mechanisms such as spot markets and ancillary services can lead to ﬁnancial gains.

Bundling power production and demand can ﬂatten a site’s power curve, which means that the size of the grid connection can be downscaled.

Advancements in terms of online communication and power system control allow an increase in the implementation of demand response programs.

Economies of scale can mean that joint, inter-ﬁrm projects can be carried out more cost-eﬃciently. This way a diverse set of technologies can be installed at a lower cost, which will lead to a more integrated power system.

It is important to note the potential of combined heat and power technology. On-site linking of the electric and thermal grid allows an increased share of renewables without jeopardizing stability [19]. Chapter 1. Introduction 6

Notwithstanding these opportunities, optimising an on-site power system is not an unambiguous task because of the multitude of involved stakeholders [9].

1.1.4 Industrial microgrids

An industrial parks power system is characterised by localised electrical and thermal loads, high energetic consumption and spatial opportunities for the integration of renewable energy sources. For these reasons, the prevailing paradigm for electric power distribution on an eco-industrial park is that of a microgrid. According to Lasseter [20], the microgrid concept assumes a cluster of loads and microsources operating as a single controllable system that provides both power and heat to the local area. The components in this energy cluster function independently in possible interaction with the surrounding macrogrid [21]. This results in enhanced local control, which can lower the cost of energy distribution, aid the integration of renewable sources and thereby reduce green house gas emissions [22]. Furthermore, the possibility of islanding improves the system security of supply by detaching the microgrid’s reliability from that of the macrogrid.

Broadly following the categorisation of the International Energy Agency [23], ﬁve kinds of ﬂexibility resources can be used to supply and balance the microgrid’s variable energetic demand: dispatchable power plants, non-dispatchable power plants, demand side management and demand response programs, energy storage facilities and interconnection with adjacent grids. A set of the most interesting technologies for a hybrid power system in Flanders is selected.

In their white paper ’A Journey to Green Energy’ [24], Flemish distribution system operator (DSO) Eandis conﬁrms that the most important renewable technologies in Belgium are and will continue to be solar and wind power. In 2015, 15% of electricity in Belgium was produced by renewable sources, with 4% of netto production by solar power and 6.5% by wind power [25]. The potential of PV and wind energy depends strongly on the solar radiation and the wind speed proﬁle at the considered site and production is therefore inherently variable in time [26].

In a power system, renewable sources with intermittent behaviour can be combined with dispatchable units in order to maintain the system balance [24]. Cogeneration, the simultaneous generation of heat and power, has a high flexibility and can therefore allow the integration of a large share of fluctuating electricity sources [19]. A combined heat and power (CHP) plant significantly increases the overall fuel efficiency compared to separate production and can

therefore lead to an important reduction in CO2 emissions [27]. At the moment the Flemish government has a certiﬁcate-based subsidy scheme for CHP installations that shows their Chapter 1. Introduction 7

interest in this technology. The share of electricity produced by cogeneration in Flanders is expected to increase [28] and therefore this technology is chosen as a component in the microgrid.

Historically, the inability to store electricity laid the foundation for the current electrical power system; the instantaneous balance between production and consumption is one of its building blocks. However, recent development in electrical storage technologies means that storing electricity has become more viable, even without large infrastructure such as a pumped hydroelectric energy storage [23]. Different aspects of microgrid flexibility (load levelling, peak shaving and seasonal storage, etc.), can therefore be offered using on-site electrical storage [29]. For these reasons, this study will assess the possibilities for electrical storage in industrial microgrid design.

Nowadays, through the introduction of the smart grid concept, the greater need for flexibility and the active participation of the end-user, demand can be seen as a new kind of resource [30]. Demand side management (DSM) is the set of methods that aims to assure grid reliability through active participation of the costumer in the electricity market [31]. Instead of changing the production side, it intents to modify the consumer’s load profile in response to financial incentives [31]. Demand management programs in a microgrid context can bring significant changes regarding economy, reliability and flexibility [32] and will therefore be included in this study.

The point of common coupling (PCC) is the connection point where the microgrid is connected to the macrogrid. While stand-alone microgrids exist [33] [34], dense electriﬁcation in Flanders means that a grid-connection can be installed at a relatively low cost. Connecting the microgrid to the utility grid makes bi-directional power exchanges possible, which allows electricity imports or exports in order to meet an excess or surplus of local production.

Figure 1.2 gives a schematic overview of the grid topology according to the technologies that were listed above. In chapter 3 each of the technologies is discussed at length and a microgrid optimisation model is developed. Chapter 1. Introduction 8

Figure 1.2: Microgrid topology Chapter 1. Introduction 9

1.2 Eiland Zwijnaarde

Eiland Zwijnaarde is an industrial park in the region of Ghent that is currently under development. It aims to accommodate a combination of high-tech companies, laboratories, university buildings and water-based logistics. It is located on an island bordered by the Ghent Canal ring in the north, the river the Scheldt in the East and the Tijarm in the south and west. An aerial view of the site is shown in Figure 1.3.

The site is being developed by Waterwegen en Zeekanaal NV and NV Eiland Zwijnaarde. NV Eiland Zwijnaarde is a collaboration between the Ghent city developer SOGent, provincie Oost Vlaanderen, and two private partners. Ghent university is expected to be one of present institutions on the park.

Figure 1.3: Map of the site Eiland Zwijnaarde

Historically the site was a dumping place for toxic waste from a nearby textile factory. At the end of last century it had become a so-called brownﬁeld and was remediated and leveled in order to prepare it for new development. Chapter 1. Introduction 10

Waterwegen en Zeekanaal NV is the owner of the site’s northern part and is planning to establish a water-based transport and logistics hub. The rest of the lands belong to NV Eiland Zwijnaarde, a public-private partnership that was created to develop the site’s southern part into a technology campus. The project is a collaboration with Ghent University, under the joint venture Tech Lane Ghent.

1.2.1 Eiland Zwijnaarde as an eco-industrial park

The City of Ghent, driven by the green-socialist coalition in power since 2012, has high ambitions regarding sustainability. By 2050 it aims to be climate neutral, which means that the entire Ghent region should have no negative impact on climate change. This is to guarantee the city’s quality of life. A climate plan (Klimaatplan) was ratiﬁed in 2015 to outline the actions that must be taken to reach this goal.

For business parks in Ghent the climate plan states that all means will be employed in order to assure sustainability, from their design until their decommissioning [35]. Speciﬁcally, reduction in energy consumption, local renewable production and rational use of energy and materials are stimulated. Furthermore, collective energy projects such as district heating will be facilitated. These sustainable principles are upheld in the development of Eiland Zwijn- aarde and the sustainability ambitions were further developed in concrete vision documents.

1.2.2 Vision of project

Two concepts play a main role in the development of the project’s vision: centralisation and sustainability. Centralisation is to what extent the ﬁrms in the park are collectively organised. Sustainability depends on how far-reaching the commitment to sustainability will be.

Two visions were developed in detail as part of project’s design process: Centralised - Maximal and Decentralised - Moderate. These are shown and explained in Figure 1.4.

Energy

In terms of energy, the two scenarios translate into the following objectives.

Central-Maximal The Central-Maximal approach aims to maximally take advantages of possible synergies between the inhabitants of the parks. The site is climate neutral and completely self- suﬃcient in terms of energy on a regional level. Employees, as well as residents of surrounding residential areas, participate in projects on sustainable energy supply. The Chapter 1. Introduction 11

Local organisation Centralised Decentralised

Intens and ambitious cooperation between Maximal public and private parties.

Sustainability

An unbinding guidance of the private partners Moderate by the public party

Figure 1.4: Development scenarios for Eiland Zwijnaarde

central park management’s list of responsibilities is extensive and many decisions on sustainability issues are taken collectively on a park level. A code of conduct makes sure that the park’s residents comply with its sustainability principles. The ambitions regarding energy are an increase of locally produced power from 20% at the site’s startup to 100% by 2050. To this end, there will be large scale wind power and solar PV, complemented with electrical storage. Later in the project a cogeneration unit can be installed that is connected to a thermal grid. BREAAM certiﬁcation (see section below) is the general guideline for the energy performance of buildings, with a focus on passive construction and maximal insulation. There is a centralised and integral energy management system that regularly carries out energy audits. The electric distribution grid is designed taking into account the renewable production and a high percentage of electric vehicles.

Decentral-Moderate For the Decentralised-Moderate scenario, the ambitions are along similar lines. However, there is less drive to carry out joint projects and the final goals are slightly less ambitious. Instead of 100% self-sufficiency by 2050, the Decentral-Moderate approach aims for 80%. Shared facilities such as storage and generation are considered but on an individual building-level instead of a park-level. A central heating network is absent, but inter-firm heat sharing between two or more firms can be considered according to the opportunities that arise. Chapter 1. Introduction 12

BREAAM certiﬁcation

The Building Research Establishment Environmental Assessment Method (BREAAM) is an evaluation method used to assess the sustainability of buildings and regions. Building and park owners, as well as city councils, governments and project developers can choose to certify their building using BREAAM in order to show their dedication to sustainability.

The certiﬁcation is not limited to energy savings and eﬃciency, instead it attempts to make a wholesome assessment according to the following categories:

Responsible management

Synergy with the surroundings

Sourcing of water, materials and energy

Spatial development

Well-being and prosperity

Regional climate and environment

Under each categories there are diﬀerent topics for which credits can be earned according to sustainability criteria. These credits are awarded according to evidence-based proof.

Trias Ecologica

An important basis for the BREAAM certiﬁcate are the Trias Ecologica and its application on energy, the Trias Energetica. Both these strategies for sustainable development were developed by Kees Duijvestein at the TU Delft.

The Trias Ecologica is a way of acting and thinking in the ﬁeld of sustainable design. The fundamental idea is to limit the input of energy, materials and water during the construction process of a building. Once it is built, the Trias aims to limit the building’s outﬂow of energy, materials and water. Concretely this means that renewable sources should be used rationally during construction, and that waste should be avoided or reused during its lifetime [11].

The application of the Trias Ecologica on the energetic design of a system results in the Trias Energetica, as shown in Figure 1.5. It consists of three strategic measures to sustainably develop and operate energy systems. The first step limits the system’s energetic consumption by eliminating losses and increasing the energetic efficiency. Secondly, consumed energy should be generated using sustainable and renewable sources. Thirdly, if non-renewable sources are inevitable, they should be used at the highest efficiency [36]. Chapter 1. Introduction 13

Reduce energy demand

Employ renewable sources Clean use of fossil fuels

Figure 1.5: Trias Energetica [36]

1.2.3 Current state of aﬀairs

In order to learn about the recent advancements in the development of Eimand Zwijnaarde, an interview with Mieke Gevaerts was conducted. She is the site’s project manager at SOGent.

The fact that both public and private stakeholders are involved in the project makes it harder to pursue a collective vision. There is political motivation to develop a sustainable industrial park that is progressive in terms of energy, mobility and ecology. However this is opposed by the very market-driven industrial park development business. Furthermore, the development process consists of many stages and in every stage each stakeholder pursues its own interests.

At the moment the project developers are trying to translate the project’s vision documents into an attractive business environment for firms. Agreements with several companies are under way and there is a large commitment from Ghent University to move to the park. As identified earlier in this chapter, active participation of the firms is a determining factor in the success of inter-firm cooperation and it is therefore important that they are involved from an early stage. However, for Eiland Zwijnaarde, the commercialisation of the project has meant that some compromises have been made on the original vision documents.

The code of conduct that would bind the companies to follow certain principles regarding sustainability will not be implemented since it would over-complicate the sales contract. In- stead, obligatory BREAAM certiﬁcation for all buildings will be used to assure sustainability. In a sales contract it is not possible to include future development regarding construction and technology. BREAAM certiﬁcation, despite its high price, has the advantage that it is updated regularly and will therefore never be outdated.

Several studies have been done on the design of the site’s energetic infrastructure. Because of the uncertain demand profiles and a possible slow ingress of companies, the proposed collective infrastructure projects such as a heating network and on-site cogeneration have been found Chapter 1. Introduction 14 economically infeasible. Instead the traditional electrical and gas grid will be installed. The potential of heat exchange with a nearby incineration facility was assessed but the fact that it can take until 2030 before the park is fully filled means that the initial thermal demand will be too low for profitability. In terms of renewable generation there are still plans for large scale wind and solar PV. At the moment one wind turbine is being certified and building-based geothermal energy is considered as a potential renewable source.

A special-purpose vehicle SPV Energie was founded to facilitate future projects regarding energy. Its role is not completely defined and can be adapted according to the site’s development. It could for example finance collective energy projects or be used as a platform for inter-firm communication on energy supply. The structure also allows participation of other stakeholders such as surrounding inhabitants or energy cooperatives. There have been talks with Energent, a Ghent-based energy cooperative, but at the moment they are not involved in the development of the site.

The different development scenarios for Eiland Zwijnaarde can be classified according to the EIP concepts defined in the previous section. In the Maximal-Central scenario, far-reaching collective measures were proposed that would affect not only the technical but also the social aspects of the park. This would lead to a very integrated industrial eco-system with strong links between the firms. The result would most closely correspond to a sustainable industrial park. The Moderate-Decentral scenario still has a focus on inter-firm cooperation but allows for more individuality. It would therefore result in an eco-industrial park. In the development scenario that is actually being carried out, the firms can choose their own role in the park’s sustainability. While the firms have to fulfil certain sustainability criteria, the final degree of sustainability will strongly depend on their own initiatives. Their motivation will be decisive as to whether Eiland Zwijnaarde will result in a green or eco-industrial park. Chapter 2

Microgrids in the Belgian electricity sector

The Belgian energy market is a complex picture with many diﬀerent actors. In order to properly assess the proﬁtability of the microgrid’s components as outlined in the section above, a general overview of the Belgian electricity sector is necessary. The aim here is to identify the roles a microgrid can play within the Belgian power sector and to assess the opportunities new technologies present in an economic context.

2.1 The liberalised electricity market

The recent transition from a state-owned, vertically integrated energy sector to a liberalised market has changed energy planning and introduced various new market roles in the electricity market. While before a single company could operate across the entire value chain of energy, diﬀerent EU liberalisation directives have disaggregated the market. Concretely this has caused a separation of the electric power business and the operation of the network infrastructure (distinguishing competitive and non-competitive parts of the industry). It allows access to energy infrastructure to third parties and implies a liberated supply side of the market with unrestricted change of supplier for the consumers [37]. Independent regulators, in Belgium the VREG, CREG, etc., were created to monitor the sector.

The aim of the transition is to lower energy prices due to increased competition and develop customised energy services. However, in the energy sector there is a long time between the initiation and completion of projects, which can be hard to reconcile with the liberalised market’s favour for high returns and quick payback times. Furthermore, for inter-ﬁrm electricity production, the strict unbundling of production and distribution as well as the prohibition of one-to-one electric connections between ﬁrms pose restrictions on the possibilities for electrical energy clustering [38].

15 Chapter 2. Microgrids in the Belgian electricity sector 16 Since the begin- 8 Technical and commercial roles in an electricity market with full retail competition ning of the new Generation Transmission

Power stations millennium, many Transmission company Sales Distribution countries have System operator Distribution companies moved towards Balance responsible Network access strong support Wholesale Spot market/PX management Metering Energy for solar and wind service provider Customer energy. Financial market (ESP) Aggregator

Ownership Money Energy Product Service

Figure 2.1: Lay-out of a liberalised electricity market [39]

9 The huge growth of wind and solar energy of recent years has created new challenges.

Figure 2.1 shows the market roles in a liberalised market. The diﬀerent actors in the electricity sector are listed below [40] [41] [42].

Production: Electricity generators are the ﬁrst players in the value chain. The power they produce is either directly consumed on-site (auto-consumption), or injected in the transmission or distribution grid. The producers sell their electricity on one of the power exchange markets or through bilateral contracts. They are in direct communication with the balance responsible party and are able to provide ancillary services to the grid according to their capacity.

Transmission: The transmission system consists of all national high-voltage lines and transmits electricity from the generators to either high-voltage (industrial) consumers or to the distribution grid. The transmission grid also allows electricity imports from abroad to satisfy national demand. In Belgium the transmission system operator (TSO) is Elia. Its role in the Belgian market is to operate the electrical system and infrastructure, which includes managing the balance between generation and consumption and maintaining the high-voltage facilities. Because of its fundamental role in the sector it acts as the market facilitator and facilitates access to the grid for producers and consumers. This energy can be provided at a price below – Distributed generation is increasing way it aims to promote free competition between the market players. It functions in a that paid by consumers on the low volt- mainly because of photovoltaics and natural monopoly and is therefore subject to monitoring by regulatory bodies. [43] age network. Because photovoltaics dis- combined heat and power and will plays an almost linear cost structure cause a significant share of genera- (without significant economies of scale in tion to be covered by a very large the investment costs) it is having a fun- number of small units. damental impact on the economics, and – Volatile production from wind and hence also the structure, of the electrici- solar energy is leading to faster and ty supply. The main characteristics of larger supply-side fluctuations of only this impact, from a technical systemic limited predictability. perspective are: – A greater geographic separation These three changes have technical im- between generation and consumption plications in all aspects of the supply and is introduced to systems previously use of electrical energy ➔ 10. Two chang- built mainly around fossil fuels or es are particularly noteworthy: the grow- nuclear energy and which previously ing importance of long-range and high- balanced consumption and genera- performance transmission networks and tion on a regional level. This develop- the integration of highly-distributed element is driven primarily by strong ments, both on the production side and location-dependent primary energy on the consumption side (smart man- sources such as wind and water. agement of consumption).

52 ABB review 4|14 Chapter 2. Microgrids in the Belgian electricity sector 17

Distribution: The distribution grid operator (DSO) operates the medium to low-voltage grid and transmits electricity from the sources to the end-users. It is responsible for the exploit- ation, maintenance and development of the distribution network and it functions just like the TSO in a natural monopoly. The DSO monitors the consumers’ consumption and passes on the information to the supplier. In Flanders diﬀerent DSOs exist such as Eandis and Infrax.

Suppliers and power exchanges: Suppliers sell electricity to households and companies. They can either buy the electricity from a power exchange or obtain power through bilateral contracts with producers. In a liberalised market suppliers can offer different types of contracts and products, depending on the costumers’ demands. Examples are ’100% renewable’ electricity, guaran- teed domestic production, etc. Suppliers are responsible for the billing of the consumed electricity. This consists of a volumetric tariff for consumed power, a fee for the network costs as well as the taxes and levies. They distribute these components to the respective market actors. They additionally charge a margin for their own operational costs. Large consumers and traders rely on the Belgian ICE power exchange to anonymously trade electricity. Different markets exist. On the Futures Market electricity can be traded 3 years ahead. About 10% of all traded electricity is sold this way. The day ahead market sells electricity for the next day with a varying price per hour. On the intra-day market electricity can be sold and purchased on a continuous basis until five minutes before delivery. Each of these markets has its own characteristics with widely varying prices that depend on the TSO zone as well as the time of the day. Market players can also decide not to trade through the power exchange and instead exchange electricity directly via bilateral contracts. This is called ’over the counter’ (OTC) or wholesale exchange. The bulk of the electricity is sold this way. OTC allows customised products to be sold and sales generally happen faster than through the power exchange. The balancing market is the market that facilitates grid balancing and is not part of the power exchange. It consists of several mechanisms to maintain the balance of the grid. These mechanisms are called ancillary services and use financial incentives to convince consumers or producers to adjust their production/consumption profile according to the grid’s needs.

Balance responsible party: The power produced at every point in time should equal the power consumed. This balance is maintained by Elia in several ways. Chapter 2. Microgrids in the Belgian electricity sector 18

At every access point in the transmission grid there are access responsible parties (ARP) appointed by Elia. They can be a producer, a consumer, an energy supplier or a trader and are responsible for the balance of power injection and extraction at that point in the transmission grid. They pay a fee for any imbalance between injections, import and purchases on the one hand, and oﬀtakes, export and sales on the other. When the ARP is unable to deliver the balancing service, the TSO can fall back on ancillary services to maintain the balance. The oﬀending ARP is charged for this.

Aggregator: Decentralised production units and loads can be grouped into a single entity when interacting with the electricity market [44]. This is called aggregation. An aggregator is a company that functions as the link between distributed energy sources and end-user on one hand and other actors such as the DSO and TSO on the other. Because of enhanced control they oﬀer, aggregators can play an important role in the transition towards a high fraction of installed non-dispatchable generation.

End-users: The end-users consume electricity to fulﬁll their needs. Consumers can be large industrial sites, oﬃce buildings, households, etc. These are either connected to the distribution grid or directly to the transmission grid. Although earlier end-users were often considered as passive consumers, at the moment there is a strong shift towards active market participation. ’Prosuming’, end-users that consume as well as produce, and demand management programs that adapt consumption according to the market’s needs, allow end-users to take an active role in the market.

2.2 Price of electricity

The price of electricity is not just a compensation for the cost of generation. The total price a consumer pays for the consumption of electricity also consists of the network costs, taxes and governmental levies. Each of the actors outlined above will have their share in the ﬁnal electricity bill. Figure 2.2 shows the breakdown in percentage of each of the diﬀerent components. The values are based on a study done by the VREG for an industrial consumer with an annual consumption of 50 MWh. Chapter 2. Microgrids in the Belgian electricity sector 19

2% 7% Commodity price

Distribution cost 41% Transmission cost 50% Levies

Figure 2.2: Breakdown of the price of electricity [45]

The commodity price and the distribution cost account for more than 90% of the total cost. It is remarkable the distribution cost is higher than the energy cost. The transmission cost is much lower than the distribution cost. VAT is not included.

2.2.1 Commodity price

Electricity is a commodity in the sense that once it is produced its source becomes irrelevant. However, the fact that large-scale electrical storage is still unproﬁtable means that electricity cannot be stored and that grid stability has a very strong impact on the instantaneous electricity price. This means that the market price of electricity can vary between one value and its tenfold within the same day, and this volatile behaviour is only aggravated by the increased integration of intermittent renewable energy sources.

In the section above the different markets in Belgium are briefly outlined. Figure 2.3 0.45 shows the Belpex day ahead market price for 0.4 the year 2015. The mean value is around 0.35 50 e per MWh of electricity, and there are 0.3 peaks of up to 450 e per MWh. This dy- 0.25 namic shows that the time of consumption 0.2 0.15 can have an strong effect on the final price 0.1 paid for electricity, and this is particularly 0.05 important for large consumers. 0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 hour

Figure 2.3: Hourly Belpex price in 2015 Chapter 2. Microgrids in the Belgian electricity sector 20

2.2.2 Transmission tarrifs

The transmission tarrif is charged by Elia and consists of the following components [46].

A. Connection The connection cost is the cost of construction for the grid connection of a speciﬁc site. It is an important parameter in this study since an increase in the microgrid’s self-suﬃciency could imply a lower connection cost. However, in Elia’s pricing scheme, most terms in the calculation of the connection cost only depend on the voltage of the connection and not explicitly on the peak power consumption[46]. The only term that does depend on the maximum power is the transformer capacity. In Elia’s calculations the transformer cost depends on its capacity according to Equation 2.1.

0.75 MVA C = C0 0.75 · (2.1) " MVA0 #

The cost of a 36kV-30kV/MT transformer with rated power 25 MVA is 34 810 e. This means that a transformer with rated power 10MVA would still cost 22 231 e. It is clear that a decrease of just 12 579 e over the lifetime of the microgrid is rather insignificant compared to the investment cost of a couple of million Euro for distributed generation units. While it is not impossible to find financial incentives for a smaller grid connection for microgrids, these will most likely not be on the transmission tariff level.

B. Management and development of grid infrastructure Cost for the monthly and yearly peak in oﬀtake and the power put at disposal.

C. Operation of the electricity system Cost for the management of the electric system and the oﬀtake of additional reactive power.

D. Compensation of imbalances Cost for power reserves, black-start and the maintenance and restoring of the residual balance of the individual access responsible parties.

E. Market integration

F. Public duties Public duties include the connection of the oﬀ-shore wind power parks in the north sea, the federal green certiﬁcates and promoting the rational use of energy, etc.

G. Taxes and levies Chapter 2. Microgrids in the Belgian electricity sector 21

In practice many of these tariffs depend on the voltage of the connection with different costs between 380/220/150 kV networks and 70/36/30 kV networks. The tariffs for networks at higher voltages is generally lower. If the yearly transmission cost for a grid connection at 30 kV is calculated for a large industrial consumer with a yearly consumption of 40 TWh and a yearly and monthly peak of 7MW, the transmission cost is 12.47 e per MWh. The main cost of the transmission is the development of the transmission infrastructure, which is 33% of the bill. It is important to note that the operational costs of the TSO, items B, C and E in the previous list, make up only half of the price paid for transmission; the cost for public duties and taxes and levies make up the other half. This shows that policy has an important impact on the total transmission bill.

2.2.3 Distribution tarrifs

The distribution tariﬀ consists of 3 aspects : a cost for power put at disposal, tariﬀs for system management and a metering cost [47]. This breakdown is analogous to the transmission cost. However, it is much harder to estimate a realistic distribution bill because the cost for distribution is much more heterogeneous depending on the geographical location of the consumer.

Just like transmission costs, distribution costs must be based on actual costs and are monitored by regulatory bodies. DSOs operating in urban regions will have much lower costs per amount of transported energy than those that operate in rural areas. In general it is possible to estimate that the distribution will be about half of the electricity bill. This is significantly more than the price paid for the transmission network, and even much higher than the commodity price paid for electricity. A decrease in the DSO’s operating costs could therefore lead to a significant decrease in the total electricity bill. While this is not the focus of this study, it would be an interesting topic of research to assess the possibilities of adapting a microgrid’s distribution grid to its loads and generators in order to find the minimal size that guarantees operation. Auto-consumption and demand shifting could be employed to minimise the use of the distribution grid and lead to a smaller investment cost.

2.2.4 Industrial energy contract

Table 2.1 gives an overview of the diﬀerent components in an industrial energy contract in Flanders. It was generated using the VREG’s VTEST tool [48] based on prices of May 2017. The prices are an estimate for an annual consumption of 10 MWh in the region Ghent with a monthly and yearly peak of 7MW. It is important to note that these prices are for a small industrial consumer connected to the low voltage grid. The speciﬁc energy cost per kWh Chapter 2. Microgrids in the Belgian electricity sector 22 gives an indication of the relative costs.

The result conﬁrms the ﬁndings of the sections above. The commodity price is around 60 e/MWh, which is a little above the 2015 mean Belpex price. The network costs are more than half of the total cost. The total price is around 240 e/MWh. Note that VAT is not included.

Table 2.1: Example of an industrial energy contract

Calculation (excl. VAT (21%)) ENERGY COST ce/kWh kWh e606.80 Annual fee 25.00 Energy component - Single day-time rate 3.78 10 000 378.00 Costs for green power 1.82 10 000 181.80 Costs for CHP 0.22 10 000 22.00 NETWORK COST e1503.58 Distribution 1331.33 Transmission 172.25 LEVIES e353.12 Contribution to energy 19.26 Federal contribution 34.11 Degressiveness Federal contribution -0.00 Contribution to the Energy Fund 299.75 TOTAL 24.64 10 000 e2463.50

2.3 Energy policy in Flanders

An overview of European climate and energy policy up to 2013 relevant to industrial energy clustering can be found in the second chapter of the ”Low carbon business park manual” written under the project ACE [11]. Energy policy in Flanders is based on the Decree on Energy (het Energiedecreet) which was drafted in 2009 and took eﬀect in 2011. The implemented provisions are described in het Energiebesluit, published in 2010 [49]. Below a brief summary of Flemish renewable energy policy and the regional measures for industry are given [50].

The federal government’s competences regarding energy consist of the transmission grid and its tariﬀs, and the production of electricity excluding renewable sources and cogeneration. Their policy for renewable energy is therefore limited to an enhanced investment deduction, which allows companies to deduct up to 13.5% (in 2017) from their taxable proﬁt for energy Chapter 2. Microgrids in the Belgian electricity sector 23 saving investments.

The Flemish government provides diﬀerent compensation schemes for energy-saving measures.

Ecology premium The ecology premium aims to stimulate companies to make their production processes energy-eﬃcient and environmentally friendly. It consists of a partial reimbursement of the made investment. Diﬀerent sustainable technologies are ranked into four categories according to their ecological impact. Financial support for a project is granted according to the investment’s category.

Strategic ecological support For investments that involve state-of-the-art technology, strategic ecological support can result in up to 40% reimbursement of the investment. It is aimed for very speciﬁc applications that are not included in the Ecology premium.

Property tax deduction For renovated and newly constructed non-residential buildings, ﬁrms can deduct a percentage from their property tax depending on the buildings E-level.

Green power certificates Electricity producers in Flanders can obtain green power certificates for electricity produced from a renewable source. Such a certificate of origin proves that a certain amount of power was produced using a renewable source and is issued by the VREG. A technology-specific banding factor determines the exact amount of energy that has to be produced to obtain one certificate. This factor depends on the evolution of the investment cost, fuel cost, etc. Certificates are awarded during the entire duration of the support period, which is generally 10 years. Green power certificates can either be sold to the DSO or to electricity suppliers. Suppliers must acquire a certain number of certificates according to an obligatory quota. Certificates, as opposed to investment support or feed-in tariffs, are less bureaucratic and allow free market mechanisms to come up with the most competitive solution.

Cogeneration certificates Cogeneration certificates are similar to green power certificates, but instead of being awarded for electricity production they are awarded per amount of primary fuel saved compared to a reference power plant and reference boiler. Just like green power certificates they can be sold to the DSO or electricity suppliers.

Support for energy audits Different schemes exist to offer firms a free and individual energy audit to identify opportunities for energy efficiency improvements. Chapter 2. Microgrids in the Belgian electricity sector 24

Subsidies and support schemes for renewables aid and accelerate the transition towards a more sustainable power system. However the large decrease in investment cost for several renewable technologies that has taken place in the past years means that these technologies no longer need additional ﬁnancial incentives to be economically interesting. Support schemes have already decreased and will continue to do so. Therefore they should not be a major consideration when planning a future power system and will not be further included in this study. Although from a developer’s point of view a decrease in subsidies could be considered as a loss of potential revenue, it is in fact a very positive trend. It means that the transition towards a sustainable power system will take place at no additional cost to society.

2.4CO 2 market

The EU emission trading system (EU ETS) was launched in 2005 to incentivise the reduction of CO2 emissions in different industrial sectors [51] . It operates according to a cap and trade principle in which companies from included industries have to compensate their emissions through emission allowances. The total annual emissions from these industries is limited and this cap reduces every year. It offers the industry flexibility in finding the most economical way of reducing their CO2 emissions. The ETS system specifically focuses on the power and heat sector because of its large share in European greenhouse gas emissions.

Allowances, which are the currency in this system, can be traded on the so-called CO2 market. They can be bought from other companies or granted by the government. While the system has been successful at eﬀectively reducing CO2 across industry [52], there is the risk of carbon leakage: businesses move their production to countries with less stringent CO2 legislation.

Furthermore, the current low carbon price (around 5 e/tCO2) means that at the moment this systems oﬀers little incentives to invest in carbon neutral technology. However, the price per tonne is expected to increase. In its study on trends in energy, transport and green house gas emissions (GHG), the European commission predicts that the ETS price will increase steeply in the coming years [53]. Their forecast is shown in Figure 2.4. This forecast will be

used in the microgrid model to economically assess the impact of a high ETS CO2 price. Chapter 2. Microgrids in the Belgian electricity sector 25 ) 2 100

80 ’13/tCO e 60

20 ETS Carbon price ( 0 2020 2030 2040 2050

Figure 2.4: Projection of ETS CO2 price

2.5 Grid structure and microgrid opportunities

The grid structure as it is now is predominantly tree shaped. Power is produced by large generating units and transported in one direction across the utility grid towards the consumer. This, however, is rapidly changing due to the global shift towards distributed generation and governmental incentives for renewable and cogeneration technologies [54]. Instead of a uni- directional power ﬂow from generator to end-user, technologies such as solar power, wind power and CHP installations allow consumers to meet their own electrical demand. This results in the need for more integrated power systems that are able to accommodate the subsequent bi-directional power ﬂows [55].

In the introduction, microgrids were identified as a technology with a large potential regarding the transition towards this new and sustainable power system. Its benefits in terms of efficiency, reliability, (cyber)security, power quality, and sustainability will positively affect the entire value chain of energy, from the producer to the end-user [56]. In a way they can be considered as fundamental building blocks of the smart grid [57]. However, there is no clear consensus on what its functionality is, how it should be controlled and what market entity should operate it. The sections below discuss these topics.

2.5.1 Functionality and economic incentives

The global trend towards regional power grids that extend beyond national borders can rein- force the advancement of microgrid technology because of its ability to integrate producers, consumers and ’prosumers’ [21]. Several types of microgrids with different functionalities can Chapter 2. Microgrids in the Belgian electricity sector 26 be identified. Campus microgrids focus on shifting away from energy intensive university facilities towards sustainably powered technology campuses [58] [59]. Because they are often located in densely electrified areas, these microgrids are generally connected to a macrogrid. Critical asset microgrids such as a hospitals [22] primarily make use of the enhanced reliability to maintain electricity supply even during macrogrid outages. Community microgrids group households and allow large civilian participation in the integration of a high RES share. Island microgrids and remote microgrids [34] [60] [33] are the most natural type of microgrids since these power systems inherently function in islanding mode and are generally not connected to a macrogrid. Microgrid technology can thus clearly be adapted to the sites’ demands regarding power quality and location specifics.

This study focuses on industrial and campus microgrids. Local control means that objectives such as optimal market participation, flattening of the load and production curves and improved power quality can be pursued [57]. To the macrogrid, an industrial microgrid is a single controlled entity, which has the advantage that is able to provide ancillary services such as black start or voltage control. The flexibility to operate on- or off-grid has a positive impact on reliability; in case of an outage on macro level, the microgrid can maintain the electrical supply by going in ’islanding’ mode. This avoids the extra cost of an individual uninterruptible power supply (UPS).

Nevertheless, a barrier for microgrids in densely electrified areas is that the additional infrastructure can be a major investment compared to a conventional grid connection. As shown in subsection 2.2.2, the low additional cost of increasing the transmission grid connection means that it is hard to economically justify a power system that allows islanding. Although the way electricity is billed at the moment makes it hard to find profitable business models for microgrid technologies, new market mechanisms such as ancillary services can offer financial support for microgrids that actively participate in the power market.

In literature the financial feasibility of microgrid projects is assessed in different cases. Us- ing the DER CAM modelling framework, Hanna et al. [22] suggest that microgrids can be cost-effective compared to complete reliance upon macrogrid utility service. A comparison is made between a large commercial building, critical infrastructure and a campus. Local production offers advantages in the case of large consumers when on-site production covers both the electrical and thermal load. According to the authors, some fraction of renewable generation will always be part of the optimal generation mix. However, due to the low gas price, cogeneration based on natural gas generators is identified as the most robust technology in the microgrid design. The authors therefore warn that without correct policies this could create difficulties in reducing green house gas emissions. Sreedharan et al. [58] specifically look at the campus of University of California - San Diego to develop a case study on the operation of its microgrid. Three roles of the microgrid in relation to the overarching utility Chapter 2. Microgrids in the Belgian electricity sector 27 grid are specifically studied: participation in the ancillary service market, peak load shifting, and on-site storage of solar power. All three are found to be technically feasible and economically profitable given the right tariff structures. Just like Hanna et al. [22], the authors emphasise the importance of integrated optimisation of both the power and heat systems. Furthermore unit commitment optimisation on monthly and daily basis is key for economic and technical efficiency. Ancillary services can be taken into account in the optimal dispatch of the units, although this can imply an increased operating cost [61]. The economic feasibility of a microgrid’s participation in the ancillary market thus strongly depends on the utility’s remuneration. Furthermore the provided services (hourly ramping, load following, frequency regulation, etc.) are determined by the microgrid’s flexibility constraints.

2.5.2 Control

Microgrid control can be divided into two hierarchical control levels: a local primary control and a smart microgrd control [57].

The local primary control is an oﬄine control system that aids the instantaneous grid stability and power quality. It functions much quicker than the secondary control. When the microgrid is in islanded mode, it is responsible for the reliability of the system. In grid connected mode it can improve power quality and voltage levels.

The high-level secondary control structure uses communication and metering to achieve technical as well as economic objectives. This is the smart control layer. It allows optimisation of the microgrid in terms of demand management, unit commitment, fuel savings and power exchange with the utility grid. It can also facilitate incentive-driven primary and secondary ancillary services to the macrogrid.

The microgrid model developed in this study only takes into account the smart layer which maintains the hourly power balance in the system. Voltage and frequency balances are not considered. Diﬀerent control strategies for the grids components were investigated and are presented in the chapter on Hybrid Power Systems. No further investigation was done regarding instantaneous grid stability.

2.5.3 Ownership

Sanz et al. [62], Hanna et al. [22] and Asmus and Lawrence [63], amongst others, identify that new business models are necessary to allow the development of microgrids. The unanswered question is: Who develops, owns, and operates them. An issue with theoretically optimised Chapter 2. Microgrids in the Belgian electricity sector 28 power systems is that because of the unbundled value chain of electricity the visions of many stakeholders have to be aligned and actual implementation of the optimal microgrid design is a complex task [64].

The following paragraphs presents a non-exhaustive lists of four organisational strategies in terms of microgrid management on industrial parks that could exist within the Belgian energy sector. For each the advantages and disadvantages are discussed.

DSO DSO-developed microgrids can become a strong case since they have plenty of experience with developing and operating power systems [63]. Because of the changes in the energy sector, DSOs need to diversify their business model and convert the power grid into an active network [38]. Developing microgrids could become a part of their new role and a step towards a smarter power system. However, one main problem with DSO operated microgrids is that legally, distribution companies are not allowed to produce or store electricity because of their natural monopoly. While they can interfere in the microgrid control through market mechanisms, this is only an indirect form of authority. This means that an additional operator is necessary. Nevertheless this barrier, because of their historic role in the electricity supply chain they will deﬁnitely become an important actor in microgrid development.

Dedicated entity A dedicated entity within the park, such as a park-specific energy supplier, is another possibility for the implementation and operation of a microgrid. Many EIPs already have a central park management to maintain facilities and to offer an inter-firm communication platform. Energy supply could either become one of this organisation’s responsibilities, or a new dedicated entity could be created. In the case of Eiland Zwijn- aarde this was done by creating SPV Energie. Setting up a cooperative energy supplier is another option. This would allow firms in the park as well as surrounding households to invest together in their own, shared electricity supply. As identified in the chapter on EIPs, this collective way of project finance has advantages in terms of responsibility and can have a positive impact on the project’s success. A drawback to this method is that diverse stakeholders are involved and that they can be hard to align towards a common goal.

External entity An external entity, such as an energy service company (ESCO), could be interested in implementing and operating an industrial park’s microgrid. ESCOs are companies whose primary business is to develop, install and ﬁnance projects regarding energy optimisation of facilities owned by their customer [65]. They often work according to Chapter 2. Microgrids in the Belgian electricity sector 29

performance-based contracts in which the revenue they receive is based on the actual energy savings due to the new technology. Since an ESCO is not bound to a single industrial park, one company could operate several microgrids and specialise in this service. Furthermore, ESCOs offer the advantage that the park’s energy management is coordinated by just one stakeholder, which offers organisational efficiency.

Individual actors Another solution could be to allow each firm to invest in their own production. The produced electricity can then be mutually traded between the other firms in the park. Recent developments regarding blockchain technology [66] means that one to one financial interactions could become quick and reliable. This way a highly interactive local power market can be constructed [67]. A remarkable case study concerns a microgrid that is currently being constructed in Brooklyn. It will use a blockchain-based ledger that will allow renewable electricity to be traded between neighbours without interme- diaries [68].

A stakeholder analysis that depends on the microgrid’s overarching organisational structure can be part of the planning process [69] [70]. This way a project can be optimised from the perspective of the diﬀerent actors. For each technology it can be decided whether a individual or collective approach oﬀers the biggest advantage. Chapter 3

Hybrid Power systems

To get familiar with the different concepts touched upon in this study on renewable generation on industrial parks, a thorough literature study was carried out on modelling of hybrid power systems. The set of technologies that was deemed promising for a microgrid in Flanders, as outlined in the introduction, are analysed. Different ways of modelling are compared and finally one is selected to be included in the microgrid simulation as used in chapter 4. For several technologies a preliminary analysis is done to see what economic conditions must be met in order to reach profitability.

The technologies that will be analysed are:

Wind power

Solar power

Combined heat and power

Electrical storage

Demand side management

Determining the optimal size of renewable generation sources is in the ﬁrst place an optimal sizing problem, and therefore the technologies above are studied from this perspective. Table 3.1 gives an overview of the 18 articles on optimal sizing of hybrid power systems that were studied. For each of these the table shows the technologies that were included in the microgrid. In chapter 4 the same list of articles is used to talk about the diﬀerent evaluation functions that can be used for the optimisation.

The developed optimisation model finds the optimal global system configuration for an industrial microgrid. However, it is important to keep in mind that in the development of an actual EIP the system will not necessary be centrally planned and each firm is responsible for their own production and demand. Each of the stakeholders will have their own interests in the optimal system design. The developed model in this study does not take this aspect into account; instead it focuses on the technical combination of technologies that offers the best performance.

30 Chapter 3. Hybrid Power systems 31

Table 3.1: Overview of technologies in studied literature

Solar Wind Battery CHP Grid DR Backup source power power storage Koutroulis et al. [71] ••• Li et al. [72] Diesel ••• Tafreshi et al. [73] Biomass •••• Fossati et al. [74] Micro turbine, •••• fuel cell & diesel Ismail et al. [75] Micro turbine ••• Zhou et al. [32] •••• Yang et al. [76] ••• Zhao et al. [33] Diesel ••• Abbes et al. [77] ••• Dufo-López and Diesel & Bernal-Agust´ın[78] ••• hydrogen Katsigiannis et al. [79] Diesel, fuel cell, ••• biodiesel & hydrogen Tito et al. [80] ••• Gonzálezet al. [81] ••• Bayod-Rújula et al. [82] •••• Theo et al. [83] Biomass •••• Kamjoo et al. [84] •••• Li et al. [85] ••• Yang et al. [86] •••

3.1 Wind power

Making an accurate model to predict wind power production at a speciﬁc site depends on wind speed data and boundary-layer meteorology and is therefore a complex task. In the section below an overview is given of the modelling techniques found in literature and the way wind power will be modelled in this study. An economic analysis is done that only includes wind power in order to learn more about the process of optimal sizing. Chapter 3. Hybrid Power systems 32

3.1.1 Wind turbine power curve

Pinst.wind Power output

0 0 Cut-in Rated Cut-out Wind speed

Figure 3.1: Wind turbine power curve

In 16 out of 18 studied articles, wind power is included as part of the hybrid energy system (see Table 3.1). All use wind speed data at the studied site to estimate the momentary wind power production.

The theoretical power in wind with speed v across an area A is calculated using equation 3.1 [84]. 1 P = ρ · A · v3 (3.1) theory 2 The wind turbine power coefficient Cp links this theoretical value with the actually produced power and depends on the specific wind turbine characteristics. Figure 3.1 shows a typical wind turbine power curve. The turbine starts producing electricity when the wind reaches the cut-in speed. As the wind speed increases, the production increases according to the curve, up to the point of rated power. For wind speeds greater than the rated wind speed, power output remains constant until the cut-out point. The wind turbine will not produce power for wind speeds above this value for safety reasons. Modelling of this curve can be done in different ways. Koutroulis et al. [71] uses pairs of wind speed and wind power production in a lookup table to determine the produced power at each time step. Other authors such as Li et al. [72], Tafreshi et al. [73] use a cubic stepwise-continuous approximation formula such as Chapter 3. Hybrid Power systems 33 the one below to determine the wind power production.

i 0, v < vci 3 avi bP , v < vi < v i  inst.wind ci r Pwind =  − (3.2) P , v < vi < v  inst.wind r co i 0, vco < v   i  With v the instantaneous wind speed, vci the cut-in wind speed, vr the rated wind speed and vco the cut-out wind speed.

Instead of the cubic correlation, some authors such as Yang et al. [76] use a linear estimation i of the power curve for wind speeds between v and v : P = P v −vci . Fossati et al. [74] ci r wind inst vr−vci i2 i uses a quadratic extrapolation method for this region: Pwind = α · v + β · v + γ.

In table 3.2 the details of the power curve are given for some examples of commercial wind turbine technologies that have been installed in on-shore wind parks in Flanders.

Table 3.2: Commercial wind turbines

Gamesa G114 Enercon E-82 GE Energy 2.5xl Rated power 2500 kW 2300 kW 2500 kW Cut-in wind speed 2,0 m/s 3,0 m/s 3,5 m/s Rated wind speed 10,0 m/s 12,0 m/s 13,5 m/s Cut-out wind speed 25,0 m/s 34,0 m/s 25 m/s

3.1.2 Wind speed data

The temporal and spatial variability and predictability of wind speeds make wind power one of the hardest renewable energy sources to predict [87]. Furthermore, hourly location- speciﬁc wind speed data is not easily available. One project that deserves mentioning is the European Meteorological derived high resolution renewable energy source generation time series (EMHIRES) project which uses the NASA atmospheric reanalysis dataset to estimate hourly wind power production over the past 30 years starting from historic wind speed data. The used NASA dataset has a time step of one hour and can therefore be used for hourly analysis [88].

According the World Meteorological Organization’s regulations, wind speed should measured at a height of 10m. To get the wind speed at hub height in order to calculate the power production from the turbine power curve, two diﬀerent extrapolation laws can be distinguished: Chapter 3. Hybrid Power systems 34

Logarithmic law: According to the logarithmic law, wind speed varies in the vertical direction according to the formula u∗ z d v = ln − + ψ(z, z0,L) (3.3) h κ z 0 with u∗ the friction velocity, κ the Von Karman constant, d is the zero plane displace-

ment, z0 is the surface roughness and ψ is a stability term. This approach is used by Kamjoo et al. [84] to calculated the wind speed at hub height. However, because it assumes neutral stability conditions it is not able to reproduce the wind speed’s expected daytime variations [89].

Exponential law: The exponential law is another extrapolation method. h a v = v (3.4) h ref h ref with a an exponent that ranges from 1/7 to 1/4. While it is a simpler method, a is not constant and depends on the height and the surface roughness [90]. Koutroulis et al. [71] and Gonz´alezet al. [81] use this law to calculate wind speeds at hub hight.

Log law 50 Power law

30 H eight [m ]

0 6 7 8 9 10 11 12 13 Wind speed [ms -1]

Figure 3.2: Wind speed proﬁles based on the log and power law

Even though much has been written on how to model wind power from wind speeds, wind speed data for the region of Flanders is hard to come by or simply does not exist. For accurate Chapter 3. Hybrid Power systems 35 data for a particular location, on-site measurements are crucial [90]. Global wind speed data sets do exist, such as MERRA which was used in the above-mentioned EMHIRES project. However, extracting data in the form of time-series from the vast map-based reanalysis data- sets is cumbersome and time consuming. Instead, the Belgian TSO Elia publishes hourly time series of wind power production and predictions since 2012 which can be downloaded from their website. Figure 3.3 shows an example for the ﬁrst week of January 2015. The aggregated dataset for on-shore wind power production in 2015 will be used for the simulations in this study. The hourly production is divided by the monitored capacity to obtain a time series of instantaneous capacity factors.

Figure 3.3: Historical wind production data

3.1.3 Dynamic economic analysis

A simulation is run to analyse the proﬁtability of wind power in more detail. The aim is to optimise the installation size, taking into account a certain load proﬁle and power exchange with the grid. It is essential to include the time component in simulations with non-dispatchable units such as wind power. This is because even though the renewable production over one year might equal the annual electric demand, big variations are possible within a smaller time step. In further simulations this method of simulation also allows to include the important dynamics of the electricity market.

The load proﬁle is that of an industrial park in the region of Ghent with an average hourly consumption of 1 MWh (8760 MWh yearly). As shown in Figure 3.4, there is no strong seasonal dependency. The installed power of the wind turbine is varied between 0 MW and 5 MW. On-shore wind power in Belgium has a capacity factor of about 0.24 which means that for Pinst.wind = 4.1 MW the yearly produced wind power matches the year electric demand Chapter 3. Hybrid Power systems 36 exactly. The hourly production, as shown in Figure 3.5, is calculated by multiplying the installed power with the hourly historical capacity factors. The power exchange from the grid i i i for every time step can then be calculated according to the formula Pgrid = Pwind LPE. i − According to this convention, power drawn from the grid Pgrid is negative and power injected in grid is positive. There is no limit on imports and exports to the grid. An example of the power curves for an arbitrary week are shown in Figure 3.7. The price for electricity exchanged with the grid is varied between 50 e/MWh and 150 e/MWh, which reﬂects the price range for an industrial consumer connected to the medium voltage grid. The buying price is 105% of this value, the selling price 95%. These value are summarised in Table 3.3

Table 3.3: Simulation parameters for wind power

Pinst.wind: 0-5 MW Consumption: 8760 MWh/year

CAPEXwind: 1900 e/kWinstalled

OPEXwind: 0.02 e/kWh Grid price: 50-150 e/MWh Life time: 25 year Interest rate: 3 %

For each value of installed power and grid price, the cost of electricity (COE) is calculated according to the formula below.

LT OPEX Total cost = CAPEX + (1 + r)j j=1 X (3.5) Total consumption = LT · Annual consumption Total cost COE = Total consumption With CAPEX the intial investment cost, OPEX the annual operating and maintenance costs, r the interest rate and LT the life time (see Table 3.3).

Figure 3.8 presents the simulation results. At a high electricity price, increasing the amount of installed wind power from 0 to 5MW will cause the COE to drop from 109.7 e to 42.2 e per MWh of consumed power. For a low grid price this trend changes; the case with no installed wind power gives a COE of 36.6 e/MWh while 5MW wind power gives 54.3 e/MWh.

These results give a good indication of the behaviour of the wind power cost function and show that even though wind power has a very low operating cost per kWh, the high initial Chapter 3. Hybrid Power systems 37

1800 3500

3000 1600 2500

2000 1400 1500

1200 1000

500

Load profile (kW) 1000 0

Power exchange with grid (kW) -500 800 -1000

600 -1500 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 hour hour

Figure 3.4: Hourly load proﬁle Figure 3.6: Hourly power exchange with the grid

4000 2000 P wind 3500 P 1500 grid LP E 3000 1000

2500 500 2000 0

1500 Power in kW

-500 1000 Wind power production (kW)

500 -1000

0 -1500 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 3000 3020 3040 3060 3080 3100 3120 3140 3160 hour hour

Figure 3.5: Hourly wind power production Figure 3.7: Power ﬂows over one week

investment cost and the necessary imported electricity for balancing make a good sizing of the installation essential. One interesting case is when the electricity price is 70 e/MWh. The simulation results for this case are presented in Figure 3.9. The COE initially decreases for increasing wind power, to reach a minimum at Pwind between 1.5MW and 2MW. If the amount of installed wind power is increased further, the COE starts to rise. This shows that an optimal value of installed wind power exists; it is exactly this point that this study aims to ﬁnd.

3.2 Solar power

Solar power has distinct characteristics compared to wind power. While it also depends on meteorological conditions, the spatial variation on a macro level is not as strong and solar Chapter 3. Hybrid Power systems 38

0.12

0.1

0.08

0.06

0.04

0.02

0.15 0.13 0.11 0.09 4000 5000 0.07 3000 0.05 2000 0 1000 P [kW] wind

Figure 3.8: 3D Simulation results for wind power production has a much more outspoken diurnal rhythm. However, clouds can cause variations on a very local scale especially during sunny days, which causes very sharp production peaks [89].

Below diﬀerent modelling techniques are outlined. An economic analysis is carried out similar to the one for wind power.

3.2.1 PV-curve & Maximum power point (MPP)

17 out of the 18 articles that are studied include solar power in their hybrid power system. In general, three main ways of modeling are identiﬁed.

The ﬁrst method used the photovoltaic (PV) current-voltage and power-voltage characteristics [71] [80]. It calculates the maximum power of the PV module depending on the solar irradiation conditions and ambient temperature according to the following formulas [71].

i · · Psolar = VOC ISC FF

VOC = VOC,ST C KV · TC − ◦ G (3.6) ISC = (ISC,ST C + KI · [TC 25 ]) · − 1000 ◦ Tnom 20 TC = T + − · G amb 800 The meteorological conditions herein are the global irradiance G and the ambient temperature

Tamb. The other variables are speciﬁcations of the photovoltaic cell: ISC is the PV short- Chapter 3. Hybrid Power systems 39

50.68 50 /MWh e

45 COE in

40 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 Pinst.wind in MW

Figure 3.9: Simulation results for grid price of 70 e/MWh

circuit current, ISC,ST C is the circuit current under standard test conditions and KI the short-circuit current temperature coefficient. VOC is the open-circuit voltage, VOC,ST C the open-circuit voltage under standard test conditions, KV the open-circuit voltage temperature coefficient. Tnom is the nominal cell operating temperature and FF the fill factor. The variables VOC and ISC are thus the temperature- and irradation-corrected output voltage and current. The derived value for the power output Psolar has to be multiplied by the number of modules connected in series and parallel to find the installation’s total power output.

A second method involves the calculation of the PV’s power output depending on the irradiation G, the area A and a temperature corrected eﬃciency:

Psolar = ηpv · A · G (3.7) ηpv = ηr · ηpc · (1 β · (Tcell Tnom)) − −

ηr is the reference module efficiency, ηpc the power conditioning efficiency and β the generator efficiency temperature coefficient. Tcell is the cell temperature and Tnom is the nominal cell temperature [77]. This method and variations thereof are used in 6 articles.

In the two methods above it is important to note that the theoretically calculated output power is the power produced under a set of meteorological conditions given that the PV module is operated in its maximum power point. Equation 3.2.1 shows the power-voltage characteristic of a PV cell under constant radiation. Depending on the external voltage that is applied to the cell’s terminals, a diﬀerent amount power is extracted. At a certain voltage

VMPP, this value reaches a maximum. Techniques for searching this maximum are called maximum power point tracking (MPPT). Chapter 3. Hybrid Power systems 40

MPP PMPP Power output

0 0 VMPP Voltage

Figure 3.10: PV power-voltage characteristic

A last way of modelling solar power production is by taking historical data directly as an indication for future production. This is done by Zhou et al. [32] and Zhao et al. [33] and will be used here. Figure 3.11 gives the solar power production in the Belgian province East-Flanders for the ﬁrst week of January 2015.

Figure 3.11: Historical solar production data

3.2.2 Dynamic economic analysis

For solar power a similar analysis is run as the one for wind. The parameters are listed in Table 3.4. Except for the capital investment cost and the operational cost they are completely Chapter 3. Hybrid Power systems 41 identical. The installed power is varied between 0 and 5MW, and the grid price is varied between 50 e and 150 e per MWh. The interaction with the grid is simulated in the same way and the COE evaluation function as well. It is important to note that since the capacity

factor of solar power in Belgium is only 12%, a total of Pinst.solar = 8.42 MW is necessary in order to produce all of the annually consumed power.

Table 3.4: Overview simulation parameters for solar power

Pinst.solar: 0-5 MW Consumption: 8760 MWh/year

CAPEXsolar: 1570 e/kWinstalled

OPEXsolar: 25 e/MWh Grid price: 50-150 e/MWh Life time: 25 year Interest rate: 3 %

A comparison between Figure 3.13 and Figure 3.7 shows that the production proﬁle of solar power is more season-dependent and therefore more predictable. A closer look at weekly power ﬂows in Figure 3.13 shows that during the night the solar production is zero. Electricity has to be imported from the grid to meet the electric demand. During the day, the production increases to reach a peak around noon, after which production starts to drop. While at night electricity is imported, during the day in general there is a surplus and electricity is sold to the grid.

4500 4000 P solar 4000 3500 P grid 3000 LP 3500 E

2500 3000 2000 2500 1500 2000 1000 Power in kW 1500 500

Solar power production (kW) 1000 0

500 -500

0 -1000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 3000 3020 3040 3060 3080 3100 3120 3140 3160 hour hour

Figure 3.12: Hourly solar power production Figure 3.13: Power ﬂows over one week

Figure 3.14 shows the 3-dimensional simulation results. As with wind power, for a high grid price an increase in the amount of installed power will cause a decrease in COE. However, in the case of solar power this decrease is not as pronounced; from 109.70 e to 93.11 e per MWh, Chapter 3. Hybrid Power systems 42 compared to 42.2 e/MWh for wind. Reasons for this are the lower power factor and the higher operational cost. For low grid prices, the COE will increase with increasing installed solar power from 36.57 e/MWh to 61.83 e/MWh.

80 79.89 0.12

0.1 /MWh e 0.08 0.06 75 0.04 COE in

0.02

0.15 0.13 5000 0.11 4000 0.09 3000 70 2000 0.07 1000 0.05 0 P [kW] solar 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 Pinst.solar in MW Figure 3.14: 3D Simulation results solar power Figure 3.15: Simulation for grid price of 110 e/MWh

An interesting case is when the grid price is 110 e/MWh. These results are shows in Fig- ure 3.15. The optimal point is reached for an amount of installed solar power between 2 and 2.5MW with a COE of 79.89 e/MWh. This example shows that solar power needs higher grid prices in order to be cost eﬃcient compared to wind power.

3.3 Cogeneration

The potential of cogeneration, or combined heat and power lies in that it is able to produce electricity and heat with higher primary energy resource eﬃciencies in comparison to separate heat and power generation [91]. An eﬃciency exceeding 90% is possible [92]. This is because the waste heat that is generated in the electricity production process is captured and reused to cover a heat demand. Figure 3.16 shows the general lay-out of a CHP installation.

There are many advantages to using cogeneration in a power system. Heavy investments in intermittent renewable power generation in recent years have created new requirements for the dynamic system reserve management of electrical grids. The short term balancing needs of power systems create opportunities for plants that can oﬀer fast system reserves. Ancillary services markets will typically reward dynamic capabilities, such as fast starts, stops and load ramps. CHP plants can proﬁtably participate in such markets provided that the power Chapter 3. Hybrid Power systems 43

Figure 3.16: Lay-out of a CHP system [93]

production can be operated independently from the heat production while maintaining high total eﬃciency at all times [94].

While CHPs are highly efficient, operate over a large load range and can provide the necessary flexibility to assist the integration of non-dispatchable sources [94], there still exist some important barriers for its complete rollout. Colmenar-Santos et al. [95] identify 3 main reasons for this trend. Primarily, as long as energy efficiency is not valued more highly in the regulatory framework, the required additional investment for CHP installations will not be attractive for utilities. Furthermore, the unharmonised legislation in Europe makes it hard for players in one national market to enter another. Secondly, the current fuel and electricity price volatility poses significant risks for CHP project developers and cause higher prices. The financial attractiveness of cogeneration is inversely proportional to the spark spread1 and as long as the spark spread keep shrinking, cogeneration projects will have a hard time to take off. Finally, the long-term nature of district heating projects is hard to reconcile with the liberalised energy market’s preference for projects with short payback periods.

3.3.1 Cogeneration technologies

Table 3.5 presents an overview of the different types of CHP installations and their performance characteristics. It can be seen that the electrical efficiency varies across a wide range, mainly depending on the size of the installation. Large systems are generally more efficient and the highest electrical efficiencies are achieved by fuel cells. While conventionally a high

1The theoretical gross margin of a gas-ﬁred power plant from selling a unit of electricity, having bought the fuel required to produce this unit of electricity. spark spread = electricity price − (gas price × heat rate) Chapter 3. Hybrid Power systems 44

electric efficiency is preferred, a lower electrical efficiency will result in more heat that can be recovered to meet the thermal demand. Overall efficiencies are more or less equal across technologies, between 70 and 80 %. The overall efficiency depends on the quality of the delivered heat.

Table 3.5: Overview of CHP technologies [96]

Recip. Engine Steam Turbine Gas Turbine Microturbine Fuel Cell

ηel 27-41% 5-40% 24-36% 22-28% 30-63%

ηtot 77-80% 80% 66-71% 63-70% 55-80%

Typical MWe 0.005-10 0.05-300 0.5-300 0.03-0.25 0.005-2 P-Q ratio 0.5-1.2 0.07-0.1 0.6-1.1 0.5-0.7 1-2 Part-load ok ok poor ok good Start-up time 10 sec 1 hr - 1 day 10 min - 1 hr 1 min sec 3 hrs - 2 days Fuels Natural gas All Natural gas, Natural gas, Hydrogen, Biogas, LPG, Synthetic gas, Sour gas, Natural gas, Industrial waste gas, Landﬁll gas, Liquid fuels Propane, Sour gas Fuel oils Methanol

Reciprocating engines Reciprocating engines have a high electrical efficiency and can be used in mobile modules. It is a mature and financially attractive technology that maintains high efficiency under part load. Its relatively high ramp rate and rapid start-up capability allows it to be used as a UPS during grid failures. However, GHG emissions tend to be high and maintenance costs as well.

Steam turbine systems Steam turbines are often bigger than reciprocating engines and can run on a wide variety of fuels. Its design is adaptable to a big range of applications. It has OK performance under part load, however its ramping time is considerably higher than for reciprocating engines. The thermal energy is of higher quality, but the power to heat ratio is low.

Gas turbines The strength of gas turbines is that they can be made in a wide range of sizes, from microturbines to large central power stations, and that the released heat is at a very high temperature. Another advantage is that GHG emissions are low. However, the electrical eﬃciency is slightly less than for reciprocating engines and steam turbines. Gas turbines are reliable and have a long life time. Chapter 3. Hybrid Power systems 45

Microturbines Microturbines are small-scale gas turbines. They are discussed separately because of their distinct operational characteristics and applications. Because micro-installation are built on-site, thermal and electrical losses can be signiﬁcantly reduced. They are mainly used for small-scale commercial and industrial applications.

Fuel cells The previous four systems all use direct combustion to convert the fuels energy first to heat and afterwards to electricity. Fuel cells do this differently; they use an electrochemical process to convert the chemical energy of hydrogen directly into electricity. Many types exist. For example, a proton exchange membrane fuel cell transforms the chemical energy liberated during the electrochemical reaction of hydrogen and oxygen to electrical energy. The recovered heat is often at a low temperature. Fuel cells can use hydrogen from different sources and the operation is characterised by the type of electrochemical process that is used. Fuel cells have a very low noise generation, low GHG emissions and perform well under part load. However, capital investment costs are high because of custom production processes.

Because of their quick start up time, reciprocating engines are chosen to be included in the microgrid model. Instead of installing a separate emergency power system, a reciprocating engine cogeneration unit combined with electrical storage can have suﬃcient capacity and ﬂexibility to replace a separate UPS.

3.3.2 Modelling and sizing

Modelling of a CHP is not unambiguous because of its two-dimensional interdependent set point, one for the thermal load and one for the electric load. Many variations in system design and control strategies make the optimal sizing of a cogeneration unit a complex task [97] and optimal sizing of a cogeneration installation in the context of microgrid design is therefore often studied separately. However, this way the interaction between dispatchable and non-dispatchable sources and the grid connection cannot be taken into account.

Only two of the mentioned authors in Table 3.1 include cogeneration in the hybrid power system’s design. Ismail et al. [75] note that very few authors on optimal sizing of hybrid power system take into account the cogeneration feature of micro-sources. The author presents a case study for a power system that includes cogeneration for remote communities. However, there is little focus on the dispatch of the CHP sources and no interaction with the utility grid. In Fossati et al. [74] the microgrid design consists of a wide range of technologies, including diﬀerent dispatchable micro-source units such as a microturbine, a fuel cell and Chapter 3. Hybrid Power systems 46 diesel generators. Heat recovered from the micro turbine and the fuel cell is used to meet the grid’s thermal demand. However, the optimisation algorithm only treats the sizing of the battery storage system. For the cogeneration units only the unit commitment is discussed.

Below an overview is given of two particular types of sizing and dispatch studies that only focus on CHP installations. The first group of articles treat the general sizing aspects and therefore do not go into much detail on the instantaneous operation. These often employ a statistical or cumulative approach rather than a time-step based simulation. The disadvantage of these methods is that they only take into account the variations of load demands, while the operation of CHP systems also depends on the instantaneous variations in fuel cost and energy policies [97]. A second set of papers is therefore studied in which detailed operational strategies are developed. These present algorithms to find the optimal operation point given an electric and thermal demand at a certain time. For each time step a fitness function is minimised; this method therefore requires substantial computational power if used for a life time analysis.

A model of a cogeneration unit is developed that is a balance between these two types; a time step based analysis that tries to emulate the optimal operation point of a cogeneration unit using simple and intuitive rules of operation.

Optimal sizing over life time: Gamou et al. [98] uses a fuel cell system to model a cogeneration system. The energy demand is simulated using a statistical approach. A hierarchical optimisation technique is used that optimises the unit size taking into account the operational planning using a penalty method. It concludes that taking into account the demand uncertainty in the design phase leads to smaller variation in the annual total cost. In Beihong and Weiding [97], a gas turbine cogeneration system is evaluated on its economic performance. The operational strategy is decided using one representative day for each month. Both the technology and the capacity of the installation are optimised using mixed-integer nonlinear programming. In Ren et al. [99] the model consists of a natural gas fired CHP plant, a thermal storage tank and a back-up boiler. It determines the optimal design capacity of these components, as well as their optimal operating strategy with the aim of minimizing the annual cost of the energy system. The capacity of CHP plant is taken as an exogenous variable that varies between 0-2kW and the power to heat ratio is assumed to be constant. A sensitivity analysis for the natural gas price, the electricity price and a carbon tax is carried out. It concludes that each project requires its own study to find the optimal installed CHP capacity, and that the results of this analysis is sensitive to the capital costs and energy prices. A thermal storage tank can make the installation more profitable as Chapter 3. Hybrid Power systems 47

long as it is not too large in order to limit heat losses. Ghadimi et al. [91] model a reciprocating engine type CHP unit in combination with an auxiliary boiler and a grid interconnection. It simplifies the operating of the installation by assuming a constant power to heat ratio and no limits on the electric operation point (0%-100%). The evaluation is based on the system’s economics, the primary energy resource consumption and environmental impact. According to the results, in order to optimise the entire energy system it is not sufficient to perform segregated sizing optimisation. Instead, integrated system sizing and operational strategy selection makes sure that the cogeneration unit works together cooperatively with the other system components. Zhang et al. [100] perform a multidimensional life-cycle analysis of four different CHP systems with an additional boiler and thermal storage unit. Three different scenarios are studied. These show the difference between one installed technology compared to multiple, and the possibility of heat dumping compared to no heat dumping. In each scenario the microgrid is connected to the utility grid to meet the electricity demand. It concludes that a microgrid with multiple CHP technologies presents a more economical and environmentally solution to meet the thermal and electric demand.

From this literature we can see that for the sizing process quite a few assumptions are made to simplify the calculations, and instead of simulating an entire year, only a few typical days are analysed. Furthermore, in order to take into account diﬀerent technologies the time component is reduced to a statistical distribution. This will result in a less accurate system evaluation. Also, short term models are not reliable to determine the optimal system conﬁguration, because more operating conditions during a much longer period should be considered for that purpose [101].

Instantaneous optimal operating point: A more complex way to model and optimize CHP based heating systems is based on a feasible operating region that is represented by a convex area determined by extreme operating points [101]. This model assumes that any point between two of the CHP’s operating points will also be a viable operating point. This is shown in ﬁgure 3.17. The marked nodes are the extreme operating points, and the shaded area is the set of all possible operating points. It is important to note that the ratio of installed

thermal power Qinst.CHP and installed electrical power Pinst.CHP depends on the selected technology. Below a certain minimum electrical load the CHP cannot operate and is shut down [101] [102]. In this study this limit is half of the installed power. Chapter 3. Hybrid Power systems 48

(P3,Q3, η3) 100

85 (P4,Q4, η4)

(P2,Q2, η2) inst.CHP % of Q (P6,Q6, η6) 20

(P1,Q1, η1) (P7,Q7, η7) 50 100

% of Pinst.CHP

Figure 3.17: CHP operating region

The cost functions, and power and heat production can then be written as

C = cjxj j∈J X P = pjxj (3.8) j∈J X Q = qjxj j∈J X with xj = 1 j∈J X (3.9) xj 0, j J. > ∈ In this way, constraints can be constructed that take into account the balances for heat, power and storages, system control, power ramping, maintenance, etc. This way of modelling is shown in [101], [103], [102] and [104]. Wang et al. [101] use a linear programming (LP) method to minimise the overall net acquisition cost for energy, looking at different CHP systems and storage scenarios. Lahdelma and Hakonen [103] also use LP but solve using a specialised Power Simplex algorithm that efficiently utilises the CHP’s specific matrix structure. In this way the dispatch problem can be solved time-efficiently on an hourly basis over the period of one year. Nazari-Heris et al. [102] compare two different CHP units by means of a multi-objective cost and emission optimisation using the -constraint method. Demand response, a fuel cell with hydrogen storage and electrical storage and a heat buffer tank are also taken into account. Alipour et al. [104]’s approach is more economic by taking into account an optimal bidding strategy, heat sales to industrial consumers and a demand response program. Non-linear power curves are modelled by Milan et al. [105]. Chapter 3. Hybrid Power systems 49

However, this way of handling optimal dispatch is too demanding in terms of computational power if the calculation is carried out over the entire life time. Therefore a compromise is found between accuracy and speed.

Developing a model: In Nazari-Heris et al. [102], the operational cost function is written as:

i i · i2 · i · i2 · i · i · i C(PCHP,QCHP) = a P CHP+b PCHP+c+d Q CHP+e QCHP+f PCHP QCHP (3.10)

i i with PCHP and QCHP the instantaneous electric and thermal set points of the installation. Example values for the coeﬃcients are shown in table 3.6

Table 3.6: CHP cost function coeﬃcients

a b c d e f 0.0435 56 12.5 0.027 0.6 0.011

200

150

100

operation 50 C

1.5 2.5 1 2 1.5 0.5 1 0.5 P [MW] 0 0 P [MW] thermal electric

Figure 3.18: Operational cost of CHP

A two dimensional plot of this function (Figure 3.18) shows a tilted ﬂat surface. This means that the marginal costs for electricity and heat are practically constant across the region of operating points. The marginal cost of heat is very low, around 0.7 e/MWh, which means that extra heat is produced at practically no extra cost. Intuitively this i means that at each set point PCHP of the produced electrical power, a certain amount i i of heat will be generated QCHP = f(PCHP). Either this heat is recovered to meet the thermal demand, or the heat is dumped if the thermal demand is smaller than the produced power; both scenarios will lead to the same fuel cost and cost for operating and maintenance. The set of operating points are therefore limited to the upper boundary Chapter 3. Hybrid Power systems 50

of the shaded region in Figure 3.17. For simplicity, the operating region is further linearised as can be seen in Figure 3.19. The thermal and electrical are also assumed

linear and shown in Table 3.7 [106]. The total eﬃciency ηtot is the sum of the separate

electric eﬃciency ηel and the thermal eﬃciency ηth.

Table 3.7: Cogeneration prime mover eﬃciencies

Electric load level ηel ηth ηtot 0.5 0.20 0.57 0.77 0.75 0.23 0.545 0.775 1 0.26 0.52 0.78

ηth = 0.52 ηel = 0.26 100

inst.CHP 60 ηth = 0.57 ηel = 0.20 % of Q

50 100

% of Pinst.CHP

Figure 3.19: Linearised CHP operating region

3.3.3 Modes of operation

For CHP installations, three general modes of operation can be identiﬁed [101]:

Constant production Depending on the size of the installed thermal power and the system’s annual heat demand, a generation unit can be made to run continuously at full load. This requires that the thermal demand profile be sufficiently season-invariant and the average thermal load be large enough to minimise heat dumping. Furthermore the installation’s electrical efficiency and the spark spread have to be taken into account in the sizing of the installation. There are different ways to limit the dumping of excess thermal and electrical energy. A bi-annual ON-OFF decision can be made based on the date; ON during Chapter 3. Hybrid Power systems 51

a climatically colder fraction of the year, OFF during the rest. Installing an absorp- tion chiller or thermal storage are other options to optimise heat usage [27].The excess electricity, for example during times of low electric load and high renewable production, can be injected in the nearby utility grid.

Electric load following Under electric load following, the installation is made to meet the system’s electric demand as far as its operating region allows. This mode of operation can be used when the microgrid’s grid connection is not so strong and self-suﬃciency is important. Auxiliary gas boilers are required to assure that the instantaneous heat demand is met. When the microgrid is acting in island mode, for example during a grid power supply failure, electric load controlled CHP installations can provide the necessary power to keep the grid stable. Ancillary services to the utility grid are also a possibility when the generator is electrically controlled and a strong grid connection is present.

Thermal load following Under thermal load following, the set point of the CHP installation is determined by the load’s heating demand. Any electricity that is produced in the process is ﬁrst used to meet the local electrical load, and excesses are injected in the nearby utility grid.

Variations and combinations of these three operating modes can be constructed depending on the characteristics of the thermal and electric load and the prices for electricity and gas. Generally the profitability of CHP installations increases with the yearly operating time. A minimum for economic viability is around 4000 to 5000 hours per year [27]. Mongibello et al. [106] compare two different operation modes, one that allows heat dumping and another that allows load partialisation. It finds that the difference between the two strategies is minimal. However, a hybrid operating strategy that takes into account the instantaneous electricity price and switches between the two modes could present a higher economic performance.

In the section below, two diﬀerent operation modes are discussed and economically evaluated. Both allow load partialisation and heat dumping. No thermal storage unit is included.

3.3.4 Dynamic techno-economic analysis

The system is modelled using the method described above, with the region of operating points as in Figure 3.19. This way each electric operating point corresponds to exactly one thermal operating point, which simplifies the model significantly. The power to heat ratio in full electric load is taken to be 2, which is a standard value for reciprocating engines. Figure 3.20 and Figure 3.21 show the industrial electric and thermal load profiles used in the simulation. Chapter 3. Hybrid Power systems 52

They are respectively based on the electric consumption of a Belgian industrial park and a generic thermal load proﬁle provided by the Flemish Energy regulator. The total annual electric consumption is 7953 MWh and the annual thermal consumption is 8000MWh.

If the heat generated by the CHP does not meet the thermal demand, an auxiliary boiler is used. If there is excess heat, this heat is dumped. Any surplus or deﬁcit of electricity is compensated using the grid connection. The capacity of both the grid connection and the auxiliary boiler is assumed to be big enough to assure that the thermal and electric load can always be met.

1600 3000

1400 2500

1200 2000

1000 1500

800 1000 Electric load profile (kW) Thermal load profile (kW)

600 500

400 0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 hour hour

Figure 3.20: Hourly electric load proﬁle Figure 3.21: Hourly thermal load proﬁle

0.45

0.4

0.35

0.3

0.25

0.2

0.15

0.1

0.05

0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 hour

Figure 3.22: Hourly Belpex price

The gas price is taken to be constant at 32.8 e/MWh and includes a fee for distribution. This price is based on the V-test by the ﬂemish energy regulator for a small industrial consumer. The boiler has an eﬃciency of 80%. Excess electricity is bought from the grid at the instant- Chapter 3. Hybrid Power systems 53 aneous Belpex price plus a fee per MWh for transmission and taxes. The Belpex price time series for 2015 is used, shown in Figure 3.22 and has a mean value of 44.7 e/MWh. The fee for transmission, distribution and taxes is 80 e/MWh; this way the average electricity price is conform with the Eurostat average electricity price for industrial consumers. Electricity is sold to the grid at the Belpex price.

In the simulations for each of the control strategies, the installed electric power is varied between 10% and 100% of the peak electric demand. The installed thermal power is each time twice that amount.

Table 3.8 gives an overview of the simulation parameters

Table 3.8: Overview simulation parameters for cogeneration case

Electric demand: 7953 MWh/year Thermal demand: 8000 MWh/year

Pinst.CHP : 0.156-1.560 MW

Qinst.CHP : 0.312-3.120 MW Electricity price: Belpex (e/MWh) Grid and taxes: 80 e/MWh Gas price: 32.8 e/MWh Life time: 25 year Interest rate: 3 %

3.3.5 System evaluation

Because in the case of cogeneration there is production of both heat and electricity, a COE per MWh (as calculated for wind and solar power) is not as easily deﬁned. For this reason the system will be evaluated using the life cycle cost (LCC), which represents the discounted total cost for covering the load’s electric and heat demands over its lifetime.

LT OPEXf,CHP + OPEXv,CHP + fuel cost + grid cost LCC = CAPEXCHP + (3.11) (1 + r)j j=1 X

The CAPEX depends on the size of the installation and is calculated using the installed electric power Pinst.CHP [91].

2 CAPEXCHP = (8e-5 · P 0.51 · Pinst.CHP + 2e3) · Pinst.CHP (3.12) inst.CHP −

The OPEX consists of a ﬁxed term OPEXf,CHP, and one term that varies OPEXv,CHP with Chapter 3. Hybrid Power systems 54

the annual produced electrical power ECHP [91].

OPEXf,CHP = 1.1e-3 · Pinst.CHP + 6.4152 − (3.13) OPEXv,CHP = ( 8e-7 · Pinst.CHP + 0.0132) · ECHP −

The amount of fuel consumed is a function of the operating points of the CHP and the auxiliary boiler. At any point in time, the fuel consumption for cogeneration is the sum of the produced electric and thermal power divided by the total system efficiency. The boiler fuel consumption is the boiler’s heat generation divided by its efficiency. The boiler’s efficiency is taken to be constant at 80%.

i i i i PCHP + QCHP Qboiler fuel = i i + ηel + ηth ηboiler 8760 (3.14) Annual fuel cost = fueli · gas price i X

The annual grid cost is calculated according to Equation 3.15.

8760 Annual grid cost = grid costi i X (3.15) P i · (Belpexi + fee),P i 0 grid costi = grid grid ≥ P i · Belpexi,P i < 0  grid grid  3.3.6 Reference case

The simulations with cogeneration are compared to a reference case without cogeneration where the entire electric demand is drawn from the grid and the entire thermal demand is generated with a gas boiler. Economic proﬁtability demands that the total energy cost with cogeneration be less than in the reference case.

The total energy cost in the reference case is 23.218 million e.

3.3.7 Electric load following:

Electric load following is implemented as described in subsection 3.3.3. The decision algorithm i i is presented in Figure 3.23. LPE and LPT are respectively the electric and thermal demand i i and PCHP and QCHP the electric and thermal operating point at time step i. Chapter 3. Hybrid Power systems 55

i i LPE, LPT

i i PCHP = QCHP = 0 yes i no i i ON LPE PI/O OFF Pgrid = LPE ≥ i i Qboiler = LPT

i PCHP = Pinst i i no i i Pinst LPE Pgrid = LPE Pinst ≥ i −i QCHP = f(Pinst)

yes i i PCHP = LPE i i i no i Pgrid = 0 QCHP < LPT Qboiler = 0 i i QCHP = f(PCHP)

yes

Qi = LP i Qi boiler T − CHP Figure 3.23: Electric load following control algorithm

For each time step the electric demand is compared to a reference value PI/O, that determines the ON-OFF behaviour of the installation. It can be written as a fraction of the installed electrical power. If the electric load is lower than the decision variable, the CHP is turned off and all electric and thermal power is supplied using the grid and the auxiliary boiler. If the load profile is higher than the decision variable, the cogeneration unit is switched on and operated to meet the electric load as far as its operating region allows. For electric loads higher than the installed electric power, the grid is used to import the additionally required power. The thermal set point is an unambiguous function of the electric set point, as shown in Figure 3.19. As a final step, the thermal generation is compared to the thermal demand and any additionally required heat is generated with the auxiliary boiler. Surplus heat is dumped.

Two variables are varied. The installed electric power Pinst.CHP is varied between 10% and

100% of the load’s peak demand. The decision variable PI/O is varied between 50% and 100% of the installed power. For each case the total energy cost is calculated. These results are Chapter 3. Hybrid Power systems 56 presented three-dimensionally in Figure 3.24.

The optimal size for the cogeneration unit lies around 40% of the peak demand, which in this case is about 624 kWe and 1248kWth. The LCC is 20.858 millione, which is below the cost for the reference case. The decision set point for this installation size is not of importance; because of the season-invariance of the electric load proﬁle (see Figure 3.20) all set points are below the minimum power demand and the CHP unit is switched on throughout the year. For higher installed powers, the total energy cost increases with an increasing value of the decision value. This conﬁrms that a higher usage of the installation increases its competitiveness.

×107

2.6

2.5

2.4

2.3 100% 90% 2.2 80% 2.1 70% 60%

2 Decision set point 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% Installed electrical power

Figure 3.24: Simulation results with electric load following

3.3.8 Thermal load following:

The algorithm for thermal load following is shown in Figure 3.25. It is analogous to electric load following, except that the IO-decision is now made according to the thermal demand and the thermal installed power. Also, the electric set-point is decided in function of the thermal set point, and not vice versa.

The optimal installed electric power is again 624 kWe (1248 kWth), however under this operating strategy the LCC is 21.386 million e, which is below the reference case but higher than the cost for electric load following. The decision variable is important even at low amount of installed power, as can be seen by the rising cost for higher decision set points. This is because the thermal load proﬁle is much more season-dependent (see Figure 3.21). Chapter 3. Hybrid Power systems 57

i i LPE, LPT

i i QCHP = PCHP = 0 yes i no i i ON LPT QI/O OFF Qboiler = LPT ≥ i i Pgrid = LPE

i QCHP = Qinst i i no i i LPT Qinst Qboiler = LPT Qinst ≤ i −i PCHP = Pinst

yes i i QCHP = LPT i i i Pboiler = 0 Pgrid = LPE PCHP i i − PCHP = f(QCHP)

Figure 3.25: Thermal load following control algorithm

×107

2.6

2.5

2.4 100%

2.3 90% 80% 2.2 70%

2.1 60% Decision set point 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% Installed electrical power

Figure 3.26: Simulation results with thermal load following Chapter 3. Hybrid Power systems 58

3.3.9 Conclusion on CHP model

The above results suggest that a control algorithm based on the combination of load partialisation and heat dumping can make a CHP installation more economical than separate heat and power production. Both electric and thermal load following result in the same amount of installed electrical capacity, which is about 40% of the peak. Of the two algorithms, electric load following results in a lower total cost, and will therefore be used in the rest of this study. The decision set point is taken to be the physical lower boundary of the installations operating region, which is half of the installed capacity.

3.4 Electrical storage

Due to the radical increase of intermittent generation units, electrical storage might prove to be crucial in the future’s electrical power grid. It can can play a set of different roles such as help manage future peak demands, allow dynamic energy management, aid the introduction of renewable power units and improve power quality [107]. Depending on the nature of the application, a different technology can be installed. Table 3.9 gives an overview of the technical characteristics of different types of electrical energy storage (EES). Three main categories can be identified: mechanical storage, electrical storage and chemical storage. Mechanical storage can store electrical energy in two ways, either through the conversion to gravitational potential energy or kinetic energy. Electrical energy storage relies on electrostatic or electromagnetic fields to store the energy. Chemical storage uses electrochemical conversion for energy storage.

While electrical storage can provide crucial services to the utility grid, its high cost is a severe obstacle for a large scale roll out. Besides the potential of electrical storage for ancillary services to the grid, long term storage and seasonal storage can mitigate the eﬀect of periods with low wind and low solar radiation on a power system with high integration of renewables. In an industrial context, storage can be economical for protection against power drops and improving power quality, but also for lowering the electricity cost through load shifting. Furthermore, storage can be part of a UPS system instead of the more conventional diesel generator [29].

In microgrids, energy storage is vital if island mode is considered and a pre-requisite for good power quality [108]. Table 3.1 shows that all but one of the studied papers include electrical storage in their hybrid power system. Even for grid-connected hybrid systems that can use the grid as a system back-up, electrical storage can be used to enhance self-consumption and reduce grid interactions to those that are economically lucrative. Chapter 3. Hybrid Power systems 59

Table 3.9: Technical parameters for diﬀerent storage technologies [29]

Round trip Depth of Lifetime in Self-discharge eﬃciency in discharge in Deployment years per hour in % % % time Now 2030 Now 2030 Now 2030 Now 2030 Mechanical: Pumped hydro 75-82 80 80-100 5e-4 3 min Compressed air 60-70 25 35-50 3e-2 5 min Flywheel 80-95 15 75 10 10 ms Electrical: Super-capacitors 90-94 15 75 0.52 < 10 ms Superconductive 80-90 20 na 0.5 magnetic Chemical: Hydrogen 34-40 40-50 40-60 6.9e-4 10 min Flow batteries 60-70 65-80 10-15 15-25 100 1.1e-2 5.2e-3 1-2 s Lithium-ion 83-86 85-92 5-20 10-30 100 6.9e-3 1.4e-3 3-5 ms Lead acid 75-80 1 5-15 10-20 70 80 1.1e-2 5.2e-3 3-5 ms High Temperat- 75-80 80-90 5-15 20-30 100 0.42 3-5 ms ure Batteries

3.4.1 Modelling

The battery is modelled as a storage device that fulﬁls three functions: to charge, to hold and to discharge energy [109]. Unlike in the case of cogeneration, there is little variation in the way electrical storage is modelled in literature. The studied papers all start from a perfect storage device with limited storage capacity and introduce diﬀerent system constraints and imperfections to derive a realistic storage model. Practically all of the studied papers use the characteristics of a lead acid battery as a starting point for their analysis.

The important variables that characterise the battery model are the following. Einst.bat is the energy capacity of the system and is the maximum amount of energy that can be stored in the system. The state of charge SOCi is how much energy the storage contains at the end

of time step i as a fraction of its maximum Einst.bat. The exchanged power with the storage i i system P exbat is how much power is either fed to the EES (P exbat > 0) or how much power i is drawn from the EES (P exbat < 0) per time step. The model assumes that all power is injected or withdrawn at the same time; at the end of the time step. Chapter 3. Hybrid Power systems 60

Table 3.10: Economic parameters for storage technologies

Power cost (e/kW) Energy cost (e/kWh) Now 2030 Now 2030 Mechanical: Pumped hydro 500-1000 5-20 Compressed air 1000 700 40-80 Flywheel 300 1000 Electrical: Super-capacitors 10-20 10000-20000 Chemical: Hydrogen 1500-2000 500-800 0.3-0.6 Flow batteries 1000-1500 600-1000 300-500 70-150 Lithium ion 150-200 35-65 300-800 150-300 Lead acid 150-200 35-65 100-250 50-80 High Temperat- 150-200 35-65 500-700 80-150 ure Batteries

In the ideal situation when no constraints are taken into consideration, the relation between the aforementioned variables is: P exi SOCi = SOCi−1 + bat (3.16) Einst.bat

Four diﬀerent types of constraints can be distinguished in literature .

Power constraint: Under the power constraint, the power drawn or stored in the battery during one time

step cannot exceed a maximum value Pinst.bat. Thus for every time step the absolute i value of the power exchange with the battery P exbat should be within a certain limit.

i 0 < P ex < Pinst.bat (3.17) | bat| This constraint is used in [74], [33] and [83], amongst others. In many cases the power constraint is part of the set of system parameters that is optimised as part of the

optimal sizing. In that way it is treated just like the installed storage capacity Einst.bat.

An increase in EES power Pinst.bat is possible but will come at a certain cost.

Eﬃciency constraint: The eﬃciency constraint takes into account the power loss that occurs when energy is converted in the storage system. This constraint is used in practically all of the studied Chapter 3. Hybrid Power systems 61

articles and for large storage systems it will be a source of signiﬁcant losses. It concretely i means that in the case of charging (P exbat > 0), the power fed to the battery is greater than the energy that is eﬀectively stored. Likewise, when a certain amount of energy i P exbat (< 0) is withdrawn from the system, only a certain fraction of this energy will be delivered to the grid.

i i−1 P exbat i SOC + ηcharge · charging (P ex < 0) i Einst.bat bat SOC = P exi (3.18) SOCi−1 + 1 · bat discharging (P exi > 0)  ηdischarge Einst.bat bat In theory the charging efficiency ηcharge and the discharging efficiency ηdischarge will have different values. However, separate values for these parameters are hard to measure [76] and battery manufacturers will therefore define a round trip efficiency for one entire charge and discharge cycle. Round trip efficiency values for different storage systems are shown in Table 3.9. In the model used here the charge and discharge efficiency are

assumed to be equal, ηcharge = ηdischarge = ηbat. ηbat is taken to be equal to the square root of the round trip eﬃciency.

ηbat = √ηroundtrip (3.19)

Some authors such as [77] and [73] assess the inverter losses separately. Here these will

be included in the system parameter ηbat.

Self-discharge: Storage systems exhibit self-discharge behaviour, which means that over time a certain fraction of the stored power will be dissipated and lost [110]. The amount of self- discharge can be expressed in percent per hour; values for diﬀerent technologies are shown in Table 3.9. It is denoted with the symbol σ. In the battery model it means that independent of the exchanged power, the SOCi at a certain time is less than the SOCi−1 of the time step before.

SOCi = (1 σ) · SOCi−1 (3.20) − Self-discharge is taken into account in [72], [73], [76], [80], [83] and [84].

Depth of discharge and maximum SOC: By introducing a maximum depth of discharge of the EES the longevity of the system can be prolonged [110] [111]. Concretely this means that when the battery is at its minimum state of charge, it no longer allows power to be withdrawn. Diﬀerent technologies have diﬀerent values for the minimum state of charge. As a constraint it is possible to write

i SOCmin < SOC < SOCmax (3.21)

SOCmax is in practice almost always equal to 1 and typical values for SOCmin are between 30% and 0%. Fossati et al. [74] identiﬁes that this limit does not take into Chapter 3. Hybrid Power systems 62

account the possible trade-oﬀ that can be made between reduced life time and instant-

aneous beneﬁt of working below the SOCmin. However, aging will not be taken into account in this work, so this trade-oﬀ cannot be evaluated.

Besides these constraints, calculating the expected lifetime of batteries is important since it can significantly influence the cost of the system [78]. Different methods can be used to determine the lifetime of the system, most of which count the cycles to failure. Furthermore, since electrical storage is not yet a mature technology, big differences can be expected in the technical as well as the economic parameters of storage systems between now and in 10 years (see Table 3.9). While generating units’ expected lifetime is at least 20 years [71], this is clearly not the case for current electrical storage technology. This model is based on a lead acid battery and the expected life time is 12.5 years. Cost estimations for batteries now and in ten years are shown in Table 3.10.

3.4.2 Battery control

Two different battery control systems were implemented. The first algorithm only takes into account the power production and consumption in the grid at each time step and tries to flatten the interaction with the grid by maximally using the battery’s capacity. The second algorithm is a fuzzy logic based control structure that takes into account the instantaneous electricity price as well as the battery’s technical constraints.

Control I - maximal usage

The first battery control strategy is based on maximally using the battery storage to limit interaction between the microgrid and the utility grid. This battery control strategy is used by Koutroulis et al. [71], Li et al. [72], Tafreshi et al. [73] , Abbes et al. [77], Tito et al. [80] and Theo et al. [83]. Figure 3.27 gives an overview of the algorithm. The input parameters are the previous state of charge SOCi−1 and the instantaneous power flow at time i, PF i. The power flow is the instantaneous difference between the electric production and the electric consumption in the microgrid at time i. This is the power that would normally be drawn from or injected in the grid. A control decision is made solely based on whether this power flow is positive or negative. If PF i > 0, there is a netto excess of electricity in the grid and power can be stored in the battery; likewise, for PF i < 0, there is a deficit of electricity that will require the battery to discharge. In both cases the magnitude

of the power ﬂow is compared to the maximum power Pinst.bat that can be drawn from Chapter 3. Hybrid Power systems 63

the battery. Any power above this value can not be compensated by the battery and will need to be exchanged with the grid. The grid interaction because of the battery’s

power limitation is called Pgrid1 and corresponds to the ﬁrst constraint discussed above. The resulting formulas for both charging and discharging are shown below. Charging Discharging

i i if: PF > P if: PF > Pinst.bat inst.bat | | i i i i Pgrid1 = PF Pinst.bat Pgrid1 = PF + ηbatPinst.bat − (3.22) (3.23) i i PF = Pinst.bat PF = Pinst.bat − The second step in the control algorithm checks to what extend the storage system’s previous state of charge allows the power transfer of PF i. This step takes into account the second, third and fourth constraints discussed above. In the case of a positive PF i (charging algorithm) the amount of vacant storage capacity is checked and ﬁlled accordingly. This can be written as:

i−1 Einst.bat i if: (SOCmax (1 σ)SOC ) · < P F − − ηbat i SOC = SOCmax (3.24) i i−1 · Einst.bat Pgrid2 = PFi (SOCmax (1 σ)SOC ) − − − ηbat

else: η PF i SOCi = (1 σ)SOCi−1 + bat − Einst.bat (3.25) i Pgrid2 = 0

For a negative PF i (discharging algorithm) the formulas are analogous. Instead of vacant capacity the constraint depends on the amount of stored capacity. i−1 i if: ((1 σ)SOC SOCmin) · ηbatEinst.bat < PF − − | | i SOC = SOCmin (3.26) i i i−1 P = PF + ((1 σ)SOC SOCmin) · ηbatEinst.bat grid2 − − else: PF i SOCi = (1 σ)SOCi−1 + − ηbatEinst.bat (3.27) i Pgrid2 = 0

The value for the total grid exchange in both charging and discharging is calculated as

Pgrid = Pgrid1 + Pgrid2. Chapter 3. Hybrid Power systems 64

SOCi−1, PF i

yes no Charge PF i > 0 Discharge

i Power i PF < Pbat PF < Pbat constraint | |

SOC SOC < SOCmax SOCmin < SOC constraint

i SOC , Pgrid

Figure 3.27: Battery control algorithm - maximal usage

While this algorithm is able to do a fair job at modeling the technical battery behaviour, it is not very eﬀective as a battery dispatch algorithm in a grid-connected microgrid. For example, when the microgrid’s own production is not so big and it depends on some constant amount of grid power for its day-to-day operation, the value of PF will be negative for every time step and the battery will never charge. Furthermore this algorithm is oblivious to economic incentives, which is important in making electrical storage cost eﬀective. It is important to note that the articles in which this strategy is implemented mainly treat stand-alone systems in which there is no grid connection. As such the battery must be used maximally in order to assure electric supply as much as possible. In a grid-connected microgrid the storage system’s objective has to be more complex. Otherwise batteries can be omitted from the hybrid system; Grid-connected systems do not necessarily require battery banks, and they mean as much as 50% of the investment cost [81]. A smart control strategy should be implemented that takes into account the variations in the grid’s electricity price. Chapter 3. Hybrid Power systems 65

Control II - fuzzy logic

For the reasons mentioned above, a fuzzy logic based control algorithm is developed that takes into account the instantaneous electricity price in the control of the battery storage. It is a simple version of the system that was developed by Fossati et al. [112] that was later used in a publication on optimal sizing by the same authors [74]. The controller’s input is the instantaneous electricity price and the power ﬂow in the

grid. From these it calculates an output Pfuzz, which is the battery’s economic set point for that time step. This value is fed to an algorithm that compares the economic set point to the battery’s physical state and it determines what power exchange can take place.

pricei, PF i

Fuzzy Control

i Charge Pfuzz Discharge

Power

i constraint i P < Pbat P < Pbat fuzz | fuzz| SOCi−1

SOC SOCi < SOC SOC < SOCi max constraint min

i SOC , Pgrid

Figure 3.28: Battery control algorithm - fuzzy logic

The strength of a fuzzy logic based control structure is that it allows to simplify decision processes that are based on multiple inputs. It does this by emulating the vagueness of a human cognitive process and thereby allows to take decisions under uncertainty [111]. Chapter 3. Hybrid Power systems 66

Fuzzy logic steps away from the crisp set theory in which an object (a number) is either a member or not a member of a certain set. Instead, it uses membership functions to define to what extend an object is part of a set [113]. This way of reasoning is intuitive because the fuzzy sets can be defined according to linguistic variables such as ”High”, ”Low”, ”Good”, ”Bad”. The output of the membership functions is a fuzzification of the original input variable. Defining a control structure with these qualitative terms is done according to linguistic rules and is therefore simple. An example of a rule is:

If Input A is ”NOT Low” and Input B is ”Very High” then Output is ”Negative”.

The interpretation of this is that the output variable belongs to the fuzzy set ”Negative” up to the point that the stated premise is true. In other words, the output is a member of the subset ”Negative” to the extend that Input A is not a member of the subset”Low” and Input B is a member of the subset ”Very High”. The combination of all these rules for a set of input variables results in a set of fuzzified output variables. The aggregated output is translated to crisp values using output membership functions [114]. In the controller developed here, the input variables are the instantaneous electricity selling price pricei and the power flow in the microgrid PF i. They are normalised using the following formulas. i i PF PFmin PFnorm = − (3.28) PFmax PFmin − pricei price pricei = − min (3.29) norm price price max − min pricemax and pricemin are respectively the highest and lowest electricity price of the year. PFmax and PFmin are the yearly highest and lowest power flow in the grid The fuzzification membership functions for the normalised inputs are shown in Fig- ure 3.29 and Figure 3.30. They are trapezoidal and the categories are ”Low” (L), ”Medium” (M), ”High” (H), Very High ”VH”. For PF there is also a membership function ”Very Low” (VL). The specific membership functions are constructed taking into account the characteristics of the normalised variables. For the electricity price, the mean of the normalised hourly Belpex price lies around around 0.1, which is the ab- scissa of the peak of the ”Medium” function. Likewise, ”VH” is chosen such that only 1% of the values during the year fall within its range. The normalised power flow is a variable that tends to be more stochastic than the electricity price, which is why the membership functions are more equally space. The linguistic control rules are displayed in Table 3.11. The main aim of this control structure is to charge the battery when the electricity price is low, and discharge when the electricity price is high. This means that the battery can charge even though Chapter 3. Hybrid Power systems 67

VL L M H VH 1

0.5 Fuzziﬁed output

0 1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1 − − − − − Crisp input

Figure 3.29: Membership functions for power ﬂow

L M H VH 1

0.5 Fuzziﬁed output

0 0 0.2 0.4 0.6 0.8 1 Crisp input

Figure 3.30: Membership functions for electricity price

the power flow in the grid is negative, or discharge when the power flow is positive. However, under no circumstances should the installed battery require fortification of the grid connection (and thus a greater grid dependency). The rules are constructed through trial and error. The basis of the control are the first i 4 rules, which solely look at the electricity price to determine Pfuzz. The final three rules make sure that for very high or very low power flow in the grid, the economic factor plays a lesser important role and technical considerations take over. This is to avoid overloading the microgrid’s grid connection. The membership functions for defuzzification are shown in Figure 3.31. The normalised

value will be multiplied by the battery power capacity Pinst.bat which is why the the range of the normalised variable is [-1 1] instead of [0 1]. Chapter 3. Hybrid Power systems 68

Table 3.11: Fuzzy control rules for battery storage

i i i price PF Pfuzz L / VP M / Z H / VN VH / VN (L) VH N ¬ (VH) VL P ¬ (H) VL P ¬

VN N Z P VP 1

0.5 Crisp output

0 1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1 − − − − − Fuzziﬁed input

Figure 3.31: Membership functions for Pfuzz

3.5 Demand-side management

Demand-side management in the context of power systems entails all the measures that have an impact on the consumption side of an energy system. While the idea of demand management is deﬁnitely not new, the current trend of distributed, volatile generation and economical communication techniques have created an environment in which dynamically adapting the consumption to the production is less expensive than vice versa [115].

Three diﬀerent categories of DSM can be identiﬁed:

Energy Eﬃciency Improving the energy eﬃciency of a power systems’ loads results in a permanent decrease in energy consumption.

Demand Response (DR) Demand Response programs shift or reduce electrical consumption, either directly or according to economic incentives, in order to ﬂatten or reduce the demand proﬁle [116]. Chapter 3. Hybrid Power systems 69

Demand response programs can be used at critical times and has the ability to both increase power grid reliability and potentially reduce operating system costs [117]. An example of a DR program is Time of Use (TOU) electricity tariffs in which a supplier charges different rates for electric consumption at different times of the day. This allows utilities to use financial incentives to convince costumers to shift their consumption from times of high demand to times of low demand.

Spinning Reserve Spinning reserve by demand management supports the utility grid by load curtailment in critical moments. It could be considered a type of demand response. According to Kirby [118], spinning reserve through demand management is the most underutilised reliability resource available to the power system today .

In this study the potential of demand response in a microgrid will be assessed. Three diﬀerent DR actions can be identiﬁed.

- Load shedding

- Load shifting to later time

- Load shifting to earlier time

Gils [119] make an assessment of demand response potential in Europe. For the belgian tertiary sector this results in a total estimate of 6337 MW of DR potential. The annual tertiary energy consumption in Belgium is 22.18TWh. This results in a DR potential of 10.4% of the average power demand.

3.5.1 Demand response modeling

Demand response is often modelled as a economic mechanism that depends on the price elasticity [31]. In this study a heuristic approach is used that looks at the daily demand proﬁle and shifts a certain percentage of the total daily energy consumption from times of high to times of low consumption. Examples for the original and resulting power ﬂow per day and per week are shown in Figure 3.32 and Figure 3.33.

Figures 3.34 and 3.35 show the annual instantaneous interaction with the grid in the case with and without demand response. It is clear that with DR the intra day variation is reduced. Table 3.12 presents the reduction in annual peak power in function of the percentage of daily shifted energy consumption. It shows that a peak reduction of around 10% corresponds to shifting 2.5 % of the load. Chapter 3. Hybrid Power systems 70

0 500 no DR without DR -200 DR 1% with DR DR 5% 0 -400

-600 -500

-800 -1000 -1000

Grid power in kW -1200 Grid power in kW -1500

-1400 -2000 -1600

-1800 -2500 2 4 6 8 10 12 14 16 18 20 22 24 0 20 40 60 80 100 120 140 160 hour hour

Figure 3.32: Demand response over one day Figure 3.33: Demand response over one week

2000 2000

1000 1000

0 0

-1000 -1000

-2000 -2000 Grid power with DR (kW) Grid power without DR (kW)

-3000 -3000

-4000 -4000 0 1000 2000 3000 4000 5000 6000 7000 8000 0 1000 2000 3000 4000 5000 6000 7000 8000 hour hour

Figure 3.34: Pgrid without DR (entire year) Figure 3.35: Pgrid with DR (entire year)

Table 3.12: Demand response simulation results

Daily shiftable load Peak load Peak reduction 0% 100% 0% 0.5% 95.3% 4.7% 1% 92.9% 7.1% 2.5% 89.3% 10.7% 5% 84.5% 15.5% Chapter 4

Optimal design and sizing

4.1 Evaluation of electricity production system

The model as described in Chapter 3 determines the dynamic behaviour of a microgrid according to its technical characteristics such as the load proﬁle, installed wind power, solar power, etc. In order to optimise this system, evaluation functions have to be deﬁned that evaluate the system’s performance.

Concretely an evaluation function f will be of the form

f(x1, x2, ..., xm) = system evaluation (4.1) with x1, x2, ..., xm etc. the technical speciﬁcations of the installation.

Table 4.1 gives an overview of the diﬀerent evaluation functions and constraints found in literature. These are explained in the section below. The abbreviations are explained in Table 4.2.

4.1.1 Economic evaluation

Life Cycle Cost (LCC): The Life Cycle Cost is the most used evaluation method in the studied literature. It is an indication of the total system cost over its lifetime, and takes into account the market’s general interest rate. The annual costs are discounted and added according to equation 4.2 [81] [71] [83].

LT Cj LCC = C + annual (4.2) initial (1 + r)j j=1 X

LT is the lifetime of the installation, Cinitial is the initial investment cost, Cannual the annual cost and r the interest rate. The initial investment cost depends on the size of the installed technologies and is shown

71 Chapter 4. Optimal design and sizing 72

Table 4.1: Overview of optimisation methods in studied literature

Evaluation Optimisation Constraints Load function algorithm Koutroulis et al. [71] LCC GA LPSP Residential Tito et al. [80] LCC GA LPSP Gonzálezet al. [81] LCC GA PB Entire township Li et al. [72] LCC GA & homer PB Island Industrial and Theo et al. [83] LCC MILP PB residential Li et al. [85] LCC Novel Tafreshi et al. [73] COE GA LPSP Remote small Ismail et al. [75] COE GA LPSP communities Yang et al. [86] COE Linear LPSP Fossati et al. [74] MOC GA PB Kamjoo et al. [84] ROI not specified PB Telecommunication Yang et al. [76] ACS GA LPSP station Bayod-Rújula et al. System none PB [82] losses Residential, Zhou et al. [32] Net profits GA Price elasticity commercial and industrial - LCC Dongfushan Zhao et al. [33] - REP GA PB Island - GHG - LCC Abbes et al. [77] - EE GA Residential home - LPSP - LCC Dufo-López and - GHG 2-layer GA PB Bernal-Agust´ın[78] - LPSP - COE - LPSP Katsigiannis et al. [79] - GHG GA - Fuel cons. - REP Chapter 4. Optimal design and sizing 73

Table 4.2: List of abbreviations and occurrence

abbrev. Evaluation function/Constraint # LCC Life cycle cost 9 COE Cost of electricity 4 ACS Annualised cost of system 1 MOC Mean operation cost 1 ROI Return on investement 1 LPSP Loss of power supply probability 9 PB Power balance 8 GHG Green house gas emissions 3 REP Renewable energy penetration 2 EE Embodied energy 1

in Equation 4.3.

Cinitial = CAPEXsolar · Pinst.solar + CAPEXwind · Pinst.wind

+2 · CAPEXc,bat · Einst.bat + CAPEXp,bat · Pinst.bat (4.3)

+CAPEXCHP · Pinst.CHP

The annual cost can include varying terms such as costs for operation and maintenance, fuel, replacement, compensation for ancillary services, decommissioning, etc. Equation 4.4 shows the formula for the annual cost used in this study. Earnings from selling power to the grid will result in a negative cost and as such decrease the LCC.

j · · Cannual = OPEXsolar Esolar + OPEXwind Ewind

+OPEXf,CHP + OPEXv,CHP · ECHP (4.4) +grid cost + fuel cost

The net present cost (NPC) and net present value (NPV) are analogous evaluation parameters, sometimes with an opposite sign convention (costs are negative and earnings positive).

Cost of Electricity (COE) The cost of electricity is used as an evaluation function in 4 of the studied articles. It is the total annual cost, as deﬁned in the calculation of the LCC, divided by the total annual consumption. Total annual cost COE = (4.5) Total annual consumption Some articles use an annualised annual cost, which is calculated by dividing the LCC by the total production over the lifetime of the installation [73]. This way interest and Chapter 4. Optimal design and sizing 74

inflation can be taken into account. LCC COE = LT (4.6) j Annual electrical consumption The COE calculates the totalP cost per unit of electricity (e.g. e/kWh), which has as an advantage that power systems of different sizes can easily be compared. However, in hybrid power systems that supply both electrical and thermal energy, the COE cannot readily be used because of the price difference between electrical and thermal energy. CHP units produce both heat and electricity from the same fuel source and it is not possible to determine exactly what fraction of fuel goes into each of these. One way to overcome this issue is by subtracting the costs of the thermal generation according to a reference scenario. This method will be used in this study.

∗ LCC Thermal reference cost COE = LT − (4.7) j Annual electrical consumption The thermal reference cost canP be calculated using a reference gas boiler with constant eﬃciency, as shown in equation 4.8 LT Annual thermal consumption gas price Thermal reference cost = · (4.8) η (1 + r)j j boiler X Annualised cost of system (ACS) The annualised cost of system is the system’s LCC, divided by its lifetime [76]. LCC ACS = (4.9) LT Mean operating cost (MOC) The mean operating cost is the sum of the mean scheduling cost and the total operating cost per day [74].

OCD = SCD + TCD (4.10)

This parameter is used when the units’ startup costs are taken into account and is too detailed to be used in this study.

Return on investment (ROI) The return on investment is another economic parameter that can be used for system evaluation. It is calculated according to the following formula [84]. TI TC ROI = − 100% (4.11) TC × With TI the total income and TC the total cost. TC consists of the sum of the inital investment cost and the present value of the maintenance cost, present value of buying electricity from the grid. It is in the ﬁrst place an investment parameter that indicates the proﬁtability of a project. It will therefore not be used in this study. Chapter 4. Optimal design and sizing 75

4.1.2 Technical evaluation and constraints

Loss of Power Supply Probability (LPSP) The loss of power supply probability is a measure of reliability for a stand-alone system [80]. It is used to assess to what extend the installation is able to cover the yearly electricity demand if there is no grid-connection to compensate instantaneous deﬁcits. It is deﬁned as in equation 4.12.

8760 Energy deﬁciti LPSP = i (4.12) i Energy demand X This evaluation method is also called Unmet load (UL) or loss of load probability (LLP). Since industrial parks are practically always grid-connected, this method is of little relevance in the evaluation of an industrial microgrid.

Power balance (PB) Power balance is the constraint that the sum of the produced and imported energy must always be equal to the sum of the consumed and exported energy. This means that no loss of power supply is allowed. It is used in all of the articles with a grid-connected microgrid [81] [82] [83] [84] and will also be implemented here.

i i i i i i Psolar + Pwind + PCHP + P exbat = Pgrid + LPE (4.13)

Embodied Energy (EE) The embodied energy is a measure for the environmental impact of the installation and is the sum of the consumed energy during the production processes of all of the system’s components [77]. comp. proc. EE = Energy demand (4.14)

The ﬁrst summation sums over theX componentsX and the second over the production processes. Values for some components can be found in literature. For other components estimations can be made by determining the relative mass of each raw material in the component. By evaluating the raw materials’ intrinsic energy and the energy requirement in the manufacturing processes a value can be derived. For the calculation of embodied energy, as with other life-cycle analysis techniques, it is very important to deﬁne a correct scope for the calculations.

Green house gas emissions (GHG) According to the IPCC’s report, the emissions of electrical generation consist of 4

factors: the direct emissions, the infrastructure and supply chain emissions, CO2 emissions due to living organisms and the albedo eﬀect and ﬁnally the emission of methane. Chapter 4. Optimal design and sizing 76

For each technology a life cycle value is calculated which represents the amount of CO2 equivalent emissions released per kWh of produced energy. This value can be multi-

plied by the produced energy per technology to ﬁnd out the total CO2 emissions. An overview of median values for diﬀerent technologies for each type of emissions are given in Table 4.3. The three values represent the minimum, median and maximum values respectively.

Table 4.3: CO2 emissions per technology in gCO2eq/kWh [120] [121]

Infrastructure & Biogenic and Direct Methane Lifecycle supply chain albedo eﬀect Gas - combined cycle 350/370/490 1.6 0 91 410/490/650 Biomass - coﬁring - - - - 620/740/890 Biomass - dedicated - 210 27 0 130/230/420 Solar power 0 42 0 0 26/41/60 Wind power - onshore 0 15 0 0 7/11/56 Grid power - - - - 102/184/262

The total CO2 emissions for the renewable sources in the microgrid over their lifetime are calculated according to:

CO2,solar = 41 · Esolar · LT

CO2,wind = 11 · Ewind · LT (4.15)

CO2,CHP = 300 · ECHP · LT

It is important to take into account that in case of a positive annual power balance with the grid (if the production in the microgrid is higher than the consumption), the

CO2 that is emitted to produce the netto export does not count towards the microgrid’s emissions. This is done by calculating the electricity mix of the installation and sub-

tracting this value from the total CO2 emissions. The value CO2,mix as calculated in

4.16 represents the average CO2 emissions per kWh of electricity produced locally in the microgrid.

41 · Esolar + 11 · Ewind + 300 · ECHP CO2,mix = (4.16) Esolar + Ewind + ECHP

If the value of Egrid is greater than zero there is a netto export to the grid and the

value of CO2,grid is negative. For Egrid < 0 there is a netto import of electricity and the

resulting CO2 emissions due to the power drawn from the grid are positive.

210 · Egrid · LT Egrid < 0 CO = − (4.17) 2,grid  CO2,mix · Egrid · LT Egrid > 0 −  Chapter 4. Optimal design and sizing 77

The total emissions are the sum of the separate components.

CO2,total = CO2,solar + CO2,wind + CO2,CHP + CO2,grid (4.18)

The relative emissions are the total emissions divided by the total consumption.

CO2,total CO2,rel. = (4.19) Econsumption · LT

Renewable energy penetration (REP) The renewable energy penetration indicates how much of the microgrid’s total energy consumption comes from renewable sources. It can be calculated according to equation 4.20 Annual local renewable production REP = (4.20) Annually consumed power This does not take into account the renewable fraction in electricity from the grid.

4.2 Optimised variables

In the model different system parameters can be optimised. These are the input variables of the evaluation functions and can be represented as a vector x1 x2 xm . The com- ··· bination of specifications with the best performance can be identifiedh using ani optimisation algorithm.

There are diﬀerent ways of deﬁning the amount of installed renewable generation. The size of the solar and wind generation can either be directly expressed in terms of installed power or in number of discrete units [71] [73] [74] [78]. For solar power these units are individual panels and for wind power these are individual turbines. Some authors decide to optimise renewable generation through the area of the PV panels [81] [77] and the swept area of the wind turbine’s rotor [77]. Additional parameters that can be optimised in the case of solar power are the tilting angle β [71] [74] and the azimuth [74]. For wind power, the optimal hub height can also be determined [71] [76].

It is possible to include technology types in the optimisation process [74] [78]. For example, a discrete decision variable could determine which type of PV technologies (monocrystaline, polycrystalline, thin ﬁlm,etc.) is installed. In the case of sizing per number, the size is of a single unit can also be included in the optimisation process.

In this study the optimisation parameters are the installed solar power, the installed wind power, the battery’s maximum power and its installed capacity and the installed electric Chapter 4. Optimal design and sizing 78 power of the CHP installation. When written as a vector it can be considered the ’DNA’ of the installation.

Pinst.solar Pinst.wind Pinst.bat Einst.bat Pinst.CHP (4.21) h i

4.3 Multi-objective optimisation

Multi-objective optimisation is the branch of mathematics that deals with the minimisation of a multidimensional evaluation function. It can aid in a multi-criteria decision making processes with conﬂicting objectives. Through specialised algorithms, an optimal set of solutions can be found.

While in single-objective optimisation the optimisation function is one-dimensional, in multi- objective optimisation it is n-dimensional (n N). Concretely this means that the evaluation ∈ function f will be of the following form.

f1(x1, x2, ..., xm) y1 f2(x1, x2, ..., xm) y2 f(x) = = (4.22) . .  .   .      f (x , x , ..., x ) y   n 1 2 m   n     The resulting system evaluation y is n-dimensional.

A solution of this problem is called nondominated or Pareto optimal if none of the objective functions can be reduced without causing an increase in some of the other functions.

4.3.1 Pareto front

An important visualisation tool in double-objective optimisation is a pareto front. It is a 2-dimensional graph with each objective presented on one of the axes. The plot consists of the pareto optimal solution set. It forms a front of which all of the points can be considered optimal; for a certain value on the x-axis the plot shows the lowest feasible value on the y-axis under the system’s constraints. The pareto front has a negative slope since an increase in one objective will lead to a decrease in the other. In this study the two objects that are minimised are the LCC and the GHG emissions. An example of a pareto curve for these evaluation functions is shown in Figure 4.1. Chapter 4. Optimal design and sizing 79

Pareto front 50

20 GHG emissions in gCO2/kWh

10 1.2 1.25 1.3 1.35 1.4 1.45 1.5 1.55 1.6 1.65 Stop Pause ×108

Figure 4.1: Graphical representation of a pareto front

4.4 Genetic Algorithm

A genetic algorithm is selected to carry out the optimisation process because of its computational strength in multi-objective optimisation [71] and because it has been used extensively in the optimisation of hybrid power systems Table 4.1.

Genetic algorithms are based on the theory of evolution and emulate the process of natural selection. The process can be divided into 4 steps. The algorithm is shown in Figure 4.2.

1. Initial population During the first step an initial population of specification vectors x is created. These vectors represents a collection of hybrid power systems with random specifications and form the basis of the optimisation process.

2. Fitness evaluation For each vector in the initial population the microgrid simulation is run and the con-

figuration is evaluated according to the predefined fitness functions (LCC and CO2 emissions). This way each member of the population is given a (multidimensional) score.

3. Selection The members that have the lowest evaluation are selected. They will be the parents of the new generation of solutions. Chapter 4. Optimal design and sizing 80

4. Genetic operators Genetic operators are used to create a new population from the group of parents.

- Selection: Members of the parents population with lowest evaluation are directly selected to be part of the next generation. - Cross-over: The DNA of multiple parents is combined to form a ’child’. - Mutation: The DNA of a single parent is slightly changed to create a new member.

5. Next generation The resulting members from the selection, cross-over and mutation operations are combined and form the next generation. The algorithm is repeated from step 2, until the diﬀerence between subsequent generations is lower than a predeﬁned value.

Genetic algorithms oﬀer a quick optimisation method for complex and multidimensional problems. However, they have the disadvantage that they can get stuck in local minima and thus arrive at a sub-optimal solution. The GA’s genetic operators can be tuned in order to mitigate this problem.

Initial population

[Psolar Pwind Pbat Ebat PCHP] Next generation

Microgrid Model Genetic operations

Fitness evaluation Selection

Figure 4.2: Flowchart of a genetic algorithm

Genetic algorithms can easily be implemented in Matlab using the ga and gamultiobj com- mands in matlab. These allow constraints to be deﬁned to which the solution vectors must comply.

Lower and upper bound Chapter 4. Optimal design and sizing 81

A lower and upper bound can be given for each element of the speciﬁcation vector x.

LBi < xi < UPi (4.23)

In this study the lower and upper bound of the system speciﬁcations will be determined by physical constraints such as available area.

Inequality constraint An inequality constraint can be imposed through the inequality matrix A and an inequality vector b. A x < b (4.24) × This can be used to for example limit the initial investment cost but will not be used here.

Equality constraint

Aeq and beq determine the equality constraint.

A x = b (4.25) × No equality constraint is deﬁned in this study. Chapter 5

Simulation results and discussion

5.1 Energy key ﬁgures and model parameters

Different studies have been done to make an estimate of the future energetic consumption of Eiland Zwijnaarde depending on the different development scenarios. Table 5.1 shows the connection values for the different scenarios as calculated by Royal HaskoningDHV and Ingenium.

Table 5.1: Predicted energetic consumption for diﬀerent scenarios

Peak demand Hours Annual demand E Base case 49 MWe 1200 58.8 GWhe Q Base case 27 MWth (Natural gas) 1500 40.5 GWh T Base case 98.8 GWh E Centralised-Maximal 50 MWe 1200 60 GWhe Q Centralised-Maximal 15 MWth (District heating) 1500 22 GWh T Centralised-Maximal 82 GWh E Decentralised-Moderate 52 MWe 1200 62.4 GWhe Q Decentralised-Moderate 15 MWth (Natural gas) 1500 22 GWh T Decentralised-Moderate 84.4 GWh E: electric consumption, Q: thermal consumption, T: total consumption

The electric peak demand is between 49 to 52 MWe, with a total annual consumption of around 60 GWhe. The thermal peak demand is between 15 and 27 MWth and the annual thermal demand lies between 22 and 40 GWh. The range of the thermal demand is much wider than for electricity because the construction and insulation methods will have a strong eﬀect on the total thermal demand.

It is important to note that the studies were ﬁrst done in 2012. The development of construction technologies and regulation can have a positive eﬀect on the buildings’ energetic performance and therefore result in a smaller thermal demand.

In the simulation the following values are used for the site’s energy consumption.

82 Chapter 5. Simulation results and discussion 83

Annual electric consumption : 60 GWh Annual thermal consumption : 40 GWh

The electric load profile is based on the consumption profile of a technology park in the region of Ghent. The thermal load profiles is a generic load profile from the Flemish Energy Regulator. They are shown in Figures 5.1 and 5.2. The peak electrical demand is 12 MW and the peak thermal demand is 14 MW.

12000 15000

11000

10000

10000 9000

8000

7000 5000 Electric load profile (kW) 6000 Thermal load profile (kW)

5000

4000 0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 hour hour

Figure 5.1: Hourly electric load proﬁle Figure 5.2: Hourly thermal load proﬁle

The upper boundary for the optimal size of the installed technologies are derived from the physical boundaries of the site. In total there is 350000 hectares available for companies. 2 At a conservative estimate of 10m /kWpeak for solar power, this means that the maximum amount of solar power that can be installed is 35 MW. For wind power it is estimated that there is space to install 3 large turbines. At 3.3 MW/turbine, the resulting upper boundary for installed wind power is 10 MW. The upper boundary for CHP is around the electric peak demand, 10 MW. There is no upper boundary for the installed battery power or energetic capacity.

The P-Q ratio of the CHP installation is 1:1. The EES has a self-discharge coeﬃcient σ of

0.001. Its eﬃciency ηbat is 90% and the minimal state of charge SOCmin is 0.15.

5.2 Sensitivity analysis

The model as described in chapter 3 is optimised for diﬀerent values of the gas and electricity price. Demand response and ETS are initially not considered. The gas price is varied between 32.8e/MWh and 122.8e/MWh and the electricity buying price between 104.6e/MWh and Chapter 5. Simulation results and discussion 84

194.6e/MWh. These prices represent a range of values for a large consumer connected to the medium voltage grid. They are between those for a residential consumer and an industrial consumer connected to the high voltage transmission grid. The electricity selling price, which is the commodity price, is equal for every year. The sensitivity analysis for the electricity price thus looks at the eﬀect of an increase in the network cost. The variation of the gas price could represent a change of fuel of the cogeneration unit. The simulations with a high gas price can in this way give an indication of the system performance when running the CHP unit on biomass.

For each combination of gas price and electricity price the model simulates the entire lifetime of the microgrid. In the ﬁrst instance the energetic demand is assumed to be constant during the life time of the installation (25 years). The values for the initial investment cost and operating cost for each technology are shown in Table 5.2. The gas price is assumed to be constant throughout the year and the electricity price varies according to the 2015 Belpex price (Figure 3.22).

Table 5.2: Economic simulation parameters [122] [29]

Battery Solar Wind CHP Capacity Power CAPEX (e/kW) 1570 1911 3940 175 175 Variable OPEX (e/kWh) 0.021 0.017 0.0108 - - Fixed OPEX (e/year) - - 9345 - -

5.2.1 System evaluation

The system is optimised according to the LCC and additionally evaluated using the COE* and the relative CO2 emissions. These evaluation forms are explained in chapter 4.

The numerical simulation results are shown in Table 1 in the appendix. Figures 5.3, 5.4 and

5.5 show 3-dimensional plots for respectively the LCC, COE* and CO2 in function of the gas and electricity price.

The LCC increases for increasing value of both the gas and electricity buying price. There is a much larger increase for an increasing gas price than for an increasing electricity price. The LCC increases by around 130 million e for a gas price increase of 90 e/MWh. An identical increase of the electricity price only causes an additional 6.78 million e in the LCC. This can be due to the fact that the increase in electricity price is modelled as an increase in only the networking cost. The microgrid’s surplus electricity is sold at the commodity price Chapter 5. Simulation results and discussion 85

which is the same for each year. As shown in Figure 5.6, in most scenarios there is a netto export of electricity. This means that electricity is only bought sporadically to maintain the instantaneous grid balance and the increased buying price therefore plays a minor role in the system cost.

×108 3 0.14

2.5 0.12

2 0.1

1.5 0.08

1 0.06 185 185 165 113 165 113 145 93 145 93 125 73 125 73 53 53 105 33 105 33

Figure 5.3: Life cycle cost Figure 5.4: Cost of electricity

The LCC plot is practically a ﬂat inclined surface, except for an irregularity at 83 e/MWh gas price and 165 e/MWh electricity price. This can be an example of the genetic algorithm that converges on a local minimum.

An increase in gas price causes a sharp drop in the CO2 emissions per kWh. An increase in

electricity price cause an increase in the CO2 emissions. As shown in Figures 5.7 and 5.8, an increase in electricity price has a minor eﬀect on the installed solar and wind power. On the other hand for a CHP installation, as shown in Figure 5.9, a higher electricity price results

in a signiﬁcant increase in the optimal installed power. The increased CO2 emissions for an increase in the electricity is thus due to the higher CHP fraction.

5.2.2 Installed power per technology

Figures 5.7, 5.8 and 5.9 show the installed power for the solar panels, the wind generation and the cogeneration unit respectively. Figure 5.10 shows the storage capacity of the battery.

For the ﬁrst three ﬁgures, there is a strong dependency on the gas price. There is a large increase of renewable generation as the gas price becomes larger. For solar power this increase is gradual and fairly uniform independent of the electricity price. For wind power the increase Chapter 5. Simulation results and discussion 86

×106 300 20 250 15 200 10

150 5

100 0 emissions in tCO2eq 2 50

CO -5 yearly grid exchange in GWh

0 185 165 33 113 185 145 93 53 165 73 145 125 73 93 125 53 113 105 105 33

Figure 5.5: CO2 emissions Figure 5.6: Power exchange with grid

is very abrupt, with a sharp rise between a gas price of 33 e/MWh and 43 e/MWh. This increase corresponds to a sharp drop in the installed CHP capacity, as shown in Figure 5.9. This shows that a slight increase in the gas price can have very strong implications on the proﬁtability of a technology.

While in all scenarios there is a high share of renewable generation, there is a limit to the physical amount of energy that can be generated on-site. Solar and wind power clearly play an important role in the optimal power system, but their share is limited by the site’s physical boundaries. This means that if the industrial park want to be completely renewable, the energy demand has to be reduced.

The installed storage capacity for the diﬀerent simulations is shown in Figure 5.10. No general trend can be found in the results. They are random variations due to the GA’s internal optimisation process. The optimised variable is the LCC, which has an order of magnitude of 100 million e. An installed battery capacity of a few hundred kWh at 175 e/kWh will therefore result in a relatively minor increase in the LCC. This allows this variation to occur. This shows that the current cost calculation method and with the implemented control strategy battery storage is not proﬁtable. Chapter 5. Simulation results and discussion 87

×104 4 10000

8000 3

6000 2 4000

1 2000 Installed wind power (kW) Installed solar power (kW) 0 0 185 185 165 113 165 113 145 93 145 93 125 73 125 73 53 53 105 33 105 33

Figure 5.7: Installed solar power Figure 5.8: Installed wind power

10000 1500

8000

1000 6000

4000 500

2000 0 Installed CHP power (kW) Installed storage capacity (kWh) 185 0 165 113 33 185 93 53 165 145 73 145 125 73 93 125 53 113 105 105 33

Figure 5.9: Installed CHP power Figure 5.10: Installed storage capacity

5.2.3 Average weekly power ﬂow

The optimal installed capacities of the microgrid’s components for an electricity selling price of 115 e/MWh and a gas price of 63 e/MWh are listed below. The size of the installed battery storage is negligible. Chapter 5. Simulation results and discussion 88

Pinst.solar : 24697 kW

Pinst.wind : 9909 kW

Pinst.CHP : 3563 kW

Einst.bat : 18kWh

Pinst.bat : 0.04kW

Figure 5.11 shows the weekly averaged power flow of the different components. The power exchange with the grid is drawn in green. During the winter there is a slight deficit in electrical power, mainly because of the reduced solar fraction. This results in a netto import of electricity from the grid (0 < Pgrid). During the summer there is a much larger production of solar which results in a netto surplus of electricity. There is a larger production of wind power during winter than during summer. However, because of the smaller installed capacity the seasonal variation in wind power is not able to compensate the solar power’s seasonal variation. The average weekly wind power exhibits high peaks between one week and the next. This results in volatile power production which is compensated by the cogeneration unit and the grid connection.

Even though there is a fairly large netto export of electricity to the grid, the CHP is con- sistently turned on throughout the year. There is not a strong seasonal dependency in the CHP’s generation, which can be explained by the fact that the CHP unit is controlled using electric load following.

8000

6000

4000

2000

Average power (kW) 0 P solar P wind P -2000 CHP P bat P grid Load profile -4000 5 10 15 20 25 30 35 40 45 50 Time (weeks)

Figure 5.11: Weekly power ﬂow during the entire year Chapter 5. Simulation results and discussion 89

×104

1.5

0.5

Average power (kW) -0.5 P solar P wind -1 P CHP P grid Load profile -1.5 20 40 60 80 100 120 140 160 Time (hours)

Figure 5.12: Hourly simulation during one week in winter

Figure 5.12 and Figure 5.13 respectively show the microgrid’s hourly production proﬁles for one week in the winter and one week in the summer.

The power ﬂows in the winter are generally much smaller than during the summer. There is much lower production of solar power which results in a power exchange with the grid that is alternately positive and negative during the day and during the night. The CHP installation is turned on most of the time. At times of low thermal load (night time) the grid connection is preferred.

During summer there are daily generation peaks due to the solar power units. These are almost directly translated into energy export peaks towards the grid. The CHP mainly works at night to compensate the absence of solar generation. There is a very clear netto positive power exchange with the grid. Chapter 5. Simulation results and discussion 90

×104

1.5

0.5

Average power (kW) -0.5 P solar P wind -1 P CHP P grid Load profile -1.5 20 40 60 80 100 120 140 160 Time (hours)

Figure 5.13: Hourly simulation during one week in summer

5.3 Future prices

Since the development of an industrial park is a time-consuming process and the prices of renewable and storage technologies are steadily decreasing, the sensitivity analysis is rerun with future cost estimates for the diﬀerent technologies. The prices are based on predictions for mid-2020’s and are summarised in Table 5.3.

Table 5.3: Future economic simulation parameters [122] [29]

Battery Solar Wind CHP Capacity Power CAPEX (e/kW) 750 1200 3940 50 50 Variable OPEX (e/kWh) 0.01 0.01 0.0108 - - Fixed OPEX (e/year) - - 9345 - -

The mean electricity buying price is varied between 0.1147 e/kWh and 0.1847 e/kWh and the gas price is varied between 0.0428 e/kWh and 0.1128 e/kWh. In total nine combinations of gas and electricity price are optimised. The numerical simulation results can be found in Table 2.

Figure 5.14 gives a plot of the LCC for the diﬀerent cases. The general shape of the curve is analogous to the results of the previous section (Figure 5.3) but there is a decrease in the Chapter 5. Simulation results and discussion 91 system cost. There is again a strong dependency on the gas price. The corresponding values for the COE are plotted in Figure 5.15. Using the future prices the COE varies between 64.9 e/MWh and 102.3 e/MWh. For the current prices this was between 86.13 e/MWh and 126.4 e/MWh. This suggests that in the future, the energy component of the billed price of electricity will most likely drop. However, the larger production peaks due to an increased share of solar and wind might result in higher network costs.

×108 2.4 0.11

2.2 0.1 2 0.09 1.8 0.08 1.6

1.4 0.07

1.2 0.06 185 185 113 113 145 145 73 73

115 43 115 43

Figure 5.14: Life cycle cost Figure 5.15: Cost of electricity

The relative CO2 per kWh is plotted in Figure 5.16. For a low gas price, CO2 emissions are constant, independent of the electricity price. There is maximum CHP production (shown in Figure 5.19) which results in fairly high emissions. As the gas price increases, the CHP fraction decreases which leads to lower emissions.

100

70 emissions in tCO2eq 2

CO 60 185 113 145 73

115 43

Figure 5.16: CO2 emissions

Figures 5.17 and 5.18 respectively show the optimal installed solar and wind power. The Chapter 5. Simulation results and discussion 92 plots show that in terms of renewable power, the price predictions will make them proﬁtable independently of the electricity and gas price. For all of the cases there is a maximal amount of installed renewable power according to the site’s physical boundaries.

The optimal battery storage capacity is shown in Figure 5.20. Again it is a seemingly random plot. This means that even for low storage prices the cost calculation and battery control algorithm do not oﬀer an economic advantage for the deployment of electrical energy storage.

×104 10000

3 8000

2 6000

4000 1 2000 Installed wind power (kW) Installed solar power (kW) 0 0 185 185 113 113 145 145 73 73

115 43 115 43

Figure 5.17: Installed solar power Figure 5.18: Installed wind power

5000 2000

4000 1500

3000 1000 2000

500 1000

Installed CHP power (kW) 0 0

185 Installed storage capacity (kWh) 185 113 113 145 145 73 73

115 43 115 43

Figure 5.19: Installed CHP power Figure 5.20: Installed batter capacity Chapter 5. Simulation results and discussion 93

5.4 Impact of ETS market

The impact of a consistent increase in the ETS CO2 market price is assessed for an electricity price of 114.7 e/MWh and a gas price of 62.8 e/MWh. The current installation costs are used as in Table 5.2.

The total amount of produced CO2 over the lifetime of the microgrid is calculated and an annual cost is paid according to the ETS price per tonne of CO2. This price is modelled to

linearly increase each year during the lifetime from 5 e/tCO2 for year 1 to 60 e/tCO2 for year 25. This is in line with the values described in chapter 4.

The resulting optimal system parameters are shown in Figure 5.21. The results of the system evaluation functions are shown below.

ETS No ETS LCC : 193 188 million e COE* : 99.5 96.1 e/MWh

CO2 : 93.564 100.24 gCO2/kWh

The LCC increases with 4.97 million e due to the extra cost for renewable production and

CO2 certiﬁcates. The COE* increases by 3.4 e/MWh.

There is a reduction in CO2 emissions of 6.68gCO2/kWh. This is a low value. The aim of the

ETS system is to aid the reduction of CO2 emissions in the EU by 80-95% by 2050 compared

to 1990 [123]. While the microgrid’s CO2 emissions per kWh are only half of the utility grid’s, it is clear that the ETS system as it is modelled here will not have a signiﬁcant additional impact to reach a further reduction.

Table 5.4 shows the optimal system conﬁguration and annually generated energy for each technology in the case with and without the ETS system.

Table 5.4: Simulation results with ETS

ETS No ETS ETS No ETS

Pinst.solar : 29.966 24.697 MW Esolar : 31186 25700 MWh/year

Pinst.wind : 9.900 9.909 MW Ewind : 21133 21150 MWh/year

Pinst.CHP : 3.567 3.563 MW ECHP : 16387 17070 MWh/year

Pinst.grid : 21.469 17.400 MW Egrid : -8696 -3925 MWh/year

The ETS system results in a higher share of solar power compared to the reference case. This requires a stronger grid connection in order to handle the large power peaks during times Chapter 5. Simulation results and discussion 94 of high solar production. The amount of wind power is at its maximal value in both cases and the CHP installation maintains its size. The netto export of power to the grid more than doubles. This can be explained by the fact that the CO2 emissions for exported power are subtracted from the grid’s total emissions; exported power therefore results in a smaller annual ETS cost.

· 104 30 ETS No ETS 3 ETS No ETS

2 20

10 0 Installed power (MW) Generated power (MWh/year) 0 -1 Solar Wind CHP Grid Solar Wind CHP Grid

Figure 5.21: Installed power with ETS Figure 5.22: Generated power with ETS

5.5 Impact of demand response

Demand response is included in the model as described in Table 5.2. On a daily basis 2.5% of the total electrical demand is shifted from times of high demand to times of low demand.

The resulting optimal installed power for each technology is shown in Figure 5.23. There is a general decrease in installed power compared to the optimisation results without demand response. The total size reduction is about 15%. Table 5.5 shows that the reduction is most pronounced for solar power and the grid connection; both decrease by more than 3MW. In terms of energy there is less solar power production and a subsequently smaller energy export to the utility grid. Chapter 5. Simulation results and discussion 95

Table 5.5: Simulation results with DR

DR No DR DR No DR

Pinst.solar : 22.216 26.402 MW Esolar : 23120 27470 MWh/year

Pinst.wind : 9.281 9.962 MW Ewind : 19810 21260 MWh/year

Pinst.CHP : 3.747 4.008 MW ECHP : 18600 17922 MWh/year

Pinst.grid : 15.204 18.732 MW Egrid : -1535 -6664 MWh/year Total installed : 50.448 59.107 MW

This shows that even a little amount of demand response can have a large impact on the installed power system. The system with DR is more self-suﬃcient without increasing the fraction of non-renewable production. There is a smaller interaction with the utility grid while maintaining a high share of renewable power.

The results of system evaluation function in the case with and without demand response are shown below.

DR No DR LCC : 193.5 191.5 million e COE* : 99.8 98.5 e/MWh

CO2 : 109.649 101.06 gCO2/kWh

Even though the system size is smaller, there is no reduction in the system cost. This is remarkable, since a smaller system size means a smaller initial investment cost. Demand reponse as it is modelled here thus increases the operational costs of the system. Several reasons can be identified that can cause this. Firstly, the demand response program is modelled as a general flattening of the demand curve. It does not respond to the instantaneous market price of electricity which means that it is possible that shifted electricity is more expensive. Secondly, the price of electricity from the grid depends in the first place on the consumed power and not on the maximum capacity. The peak power of the grid connection has no large impact on the total price, which is why a reduction in grid connection is not favourable. A different pricing scheme could make demand response economically more interesting. Chapter 5. Simulation results and discussion 96

· 104 30 DR No DR 3 DR No DR

2 20

10 0 Installed power (MW) Generated power (MWh/year) 0 -1 Solar Wind CHP Grid Solar Wind CHP Grid

Figure 5.23: Installed power with DR Figure 5.24: Generated power with DR

5.6 Impact of gradual ingress

In the section on Eiland Zwijnaarde it was identified that an unpredictable energetic demand during the first years of the park is a limiting factor for the development of shared energy facilities. The additional investment cost for technologies such as a thermal network was said to be too high if the energetic demand is not sufficiently large. The developed model is used to asses this claim and to quantify the impact of a variable annual energetic demand on the optimal configuration.

In the previous sections the hybrid power system was sized according to an equal energetic demand for every year of its lifetime. However, a gradual ingress of companies on the site will result in a much smaller load profile during the first years of the park. In the model, the first year’s electric and thermal demand are set to 10% of the nominal value. During the following ten years the demands increase by 10% every year, as shown in Figure 5.25. After 10 years the electric and thermal demand are at their nominal value.

Nominal annual electric consumption : 60 GWh Nominal annual thermal consumption : 40 GWh Chapter 5. Simulation results and discussion 97

100%

75%

50% Energetic demand 25%

1 5 10 15 20 Year

Figure 5.25: Eﬀect of ingress on energetic demand

The power system is sized once and the installed capacities are ﬁxed during the entire lifetime of the installation. In the case of non-dispatchable units this means that the yearly production is constant without regard to the varying demand. For the dispatchable units the unit commitment algorithm is run separately for each year with diﬀerent demand.

The system is optimised for an average electricity buying price of 134.7 e/MWh and a gas price of 62.8 e/MWh. The installation cost are the current market prices as shown in Table 5.2.

The optimisation results in terms of installed power and yearly generated electricity are shown in Figures 5.26 and 5.27. Table 5.6 gives the installed power and generated electricity in the case with gradual ingress and a reference case without gradual ingress.

Table 5.6: Simulation results with gradual ingress

Ingress Constant Ingress Constant

Pinst.solar : 8.130 26.402 MW Esolar : 8461 27477 MWh/year

Pinst.wind : 9.872 9.962 MW Ewind : 21072 21264 MWh/year

Pinst.CHP : 0 4.008 MW ECHP : 0 17922 MWh/year

Pinst.grid : 12.966 18.732 MW Egrid : 30466 -6664 MWh/year Chapter 5. Simulation results and discussion 98

· 104 30 Ingress No ingress 3 Ingress No ingress

2 20

10 0 Installed power (MW) Generated power (MWh/year) 0 -1 Solar Wind CHP Grid Solar Wind CHP Grid

Figure 5.26: Installed power with ingress Figure 5.27: Generated power with ingress

There is a clear absence of cogeneration in the optimised system. Instead, all of the thermal energy is provided by the auxiliary boiler. The CHP unit is controlled according to the electrical demand, which means that during the ﬁrst year it would practically never be turned on. Its absence has as a results that the electrical power mainly comes from the grid and the wind turbines. While in the previous optimisations there was always a netto export of electricity to the grid, in this case there is a large netto import. This grid dependency does not result in a larger grid connection; the grid’s installed power decreases from around 18 MW to 13 MW. This is due to the lower installed solar fraction. Solar power gives large production peaks during summer which require a strong grid connection to inject the energy surplus on the grid. Decreasing the solar fraction thus means that the grid connection can be made smaller.

The cost of electricity is 108.2 e per MWh. This is about 10 e higher per MWh compared to the reference scenario.

The weekly average power ﬂow for the fourth and the tenth year are shown in Figure 5.28. The solar and wind power production are in both cases equal. There is a large increase in the annual demand from year 4 to year 10. All of this energy is compensated using the grid connection. In the beginning of the grid’s lifetime there is a large injection of renewable power to the utility grid. After 10 year this has changed and there is a netto import of electricity.

Figure 5.29 shows the hourly power curves during one week for four diﬀerent years. In the ﬁrst year there is very little demand, which means that practically all the produced electricity is injected in the grid. In the subsequent years, the load increases and power injection to the Chapter 5. Simulation results and discussion 99

Year 4 Year 10-25 8000 8000

6000 6000

4000 4000

2000 2000

0 0 Average power (kW) Average power (kW) P -2000 -2000 solar P wind P grid -4000 -4000 Load profile

5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 Time (weeks) Time (weeks)

Figure 5.28: Weekly power ﬂow for the fourth and tenth year

utility strongly decreases. Note that the plots show one week during summer, which means that the capacity factor for solar power is relatively high.

The high injection peaks in the early years of the park result in a grid connection that is over-sized for the nominal load. In the ﬁrst year the peak power exchange with the grid is almost 13 MW, while during the years 10-25 this is only 9.3 MW. This tempor- ary large imbalance between production and consumption can be avoided by considering the sizing of the power system as a dynamic process. Instead of installing the entire production at once, the renewable production can grow gradually according to the size of the demand. This can be implemented in the model by replacing the optimisation vector

x = Pinst.solar, Pinst.wind Pinst.bat Einst.bat Pinst.CHP with an optimisation matrix x. h i

Pinst.solar,1 Pinst.solar,2 Pinst.solar,nom ··· Pinst.wind,1 Pinst.wind,2 Pinst.wind,nom  ···  x = Pinst.bat,1 Pinst.bat,2 Pinst.bat,nom (5.1)  ···     Einst.bat,1 Einst.bat,2 Einst.bat,nom   ···  Pinst.CHP,1 Pinst.CHP,2 Pinst.CHP,nom  ···    th Ptech,j represents the installed capacity of a certain technology at the j step of the devel-

opment process. Ptech,nom is the nominal installed capacity when the park is fully developed. This was not further investigated in this study but could be an interesting topic for future research. Chapter 5. Simulation results and discussion 100

×104 Summer - year 1 ×104 Summer - year 4 1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4 Average power (kW) Average power (kW) P P -0.6 solar -0.6 solar P P -0.8 wind -0.8 wind P P CHP CHP P P -1 grid -1 grid Load profile Load profile -1.2 -1.2 20 40 60 80 100 120 140 160 20 40 60 80 100 120 140 160 Time (hours) Time (hours)

×104 Summer - year 7 ×104 Summer - year 10-25 1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4 Average power (kW) Average power (kW) P P -0.6 solar -0.6 solar P P -0.8 wind -0.8 wind P P CHP CHP -1 P -1 P grid grid Load profile Load profile -1.2 -1.2 20 40 60 80 100 120 140 160 20 40 60 80 100 120 140 160 Time (hours) Time (hours) Figure 5.29: Hourly power curves over one week in summer for diﬀerent years

5.7 Multi-dimensional analysis

Using the gamultiobj function in matlab the nondominated set of optimal system conﬁgur-

ations is determined according to the LCC and the relative CO2 emissions. The simulation is carried out for a mean electricity price of 114.6 e/MWh and a gas price of 62.8 e/MWh with energetic demand and component costs as in Table 5.2. The pareto plot is shown in Figure 5.30. Chapter 5. Simulation results and discussion 101

Pareto front 100

GHG emissions in gCO2/kWh 60

50 1.88 1.9 1.92 1.94 1.96 1.98 2 2.02 2.04 Stop Pause ×108

Figure 5.30: 2-dimensional pareto front

The conﬁguration with the lowest LCC and the highest CO2 lies around 18.8 million e with

emissions of 96gC02/kWh. This corresponds to the result of the single-objective optimisation

carried out in section 5.2. The CO2 emissions initially drop rapidly for an increasing value of

LCC. A reduction of 8gCO2/kWh comes at an additional cost of 949000 e. On average this is

a cost of 117000 e per gCO2/kWh. For an LCC greater than 19 million e with emissions below

86gC02/kWh the curve is practically linear. In this region the cost of saving an additional

gC02/kWh is equal to about 340000 e.

×104 ×107 3.5 3.5

3 3 P solar P 2.5 wind 2.5 P CHP 2 P E grid solar 2 E 1.5 wind E CHP 1 E 1.5 grid

0.5 Installed power (kW)

1 Generated power (kWh) 0

0.5 -0.5

0 -1

95.6gCO /kWh 71.74gCO /kWh 53.94gCO /kWh 95.6gCO /kWh 71.74gCO /kWh 53.94gCO /kWh 2 2 2 2 2 2

Figure 5.31: Installed power on pareto front Figure 5.32: Generated power on pareto front

Figures 5.31 and 5.32 respectively show each technology’s installed power and annually gen- Chapter 5. Simulation results and discussion 102

erated electricity for every point on the pareto front. For low LCC, the reduction in CO2 is due to a decrease in cogeneration and an increase in solar power. This corresponds to the steepest part of the pareto curve which means that this reduction comes at the lowest cost. The wind power fraction and grid export stay relatively constant. For increasing LCC,

the installed solar power reaches its maximum. For a further CO2 reduction the installed power of the CHP keeps decreasing and the energy deﬁcit is compensated by less electricity

exports to the grid. This way the emissions can be reduced down to about 54 gCO2/kWh. In comparison, the average emissions per kWh of electricity in Belgium is between 180 and

210 gCO2 [121]. In Europe this is around 320 gCO2/kWh [124]. The netto power exchange

with the grid becomes positive for very low values of CO2 Chapter 6

Conclusion

This study assesses the techno-economic opportunities of an industrial microgrid in the design and operation of an eco-industrial park.

A literature study on eco-industrial park design is carried out in order to identify the drivers and limitations of an inter-firm energy supply. The result is a multidisciplinary list of recommendations for park owners and developers. The main advantages for a collective power system is the possibility of bundling power production and demand in order to flatten the site’s load profile. Furthermore, economies of scale make that joint projects are generally cheaper per installed kW. Combined heat and power is identified as a technology with a lot of potential, especially towards the integration of a large share of on-site renewable energy sources. Ideally, projects on shared utilities are carried out with an active participation of the park’s firms. A drive towards sustainability can ideologically be stimulating for a project but the financial gain is what will be the main driver.

The concepts of a collective power system are applied to the case of Eiland Zwijnaarde, an industrial park that is currently being developed in the region of Ghent. The park developers have strong sustainability ambitions, but their vision turns out to be hard to translate to reality. Initially the aim was to create an integrated sustainable business park with large shared energetic facilities and a high level of collectivity. Uncertainty about the energetic demand and the importance of profitability in industrial park development have resulted in a current development plan that allows the firms more freedom in their sustainability choices. BREAAM certification will be the main tool to make sure that the sustainability criteria are fulfilled. Projects on collective and renewable energy supply are facilitated by the SPV Energie that will provide a platform for inter-firm cooperation. Self-organised initiatives between firms are encouraged and will offer advantages in terms of continuation and ownership of the projects.

A brief overview of the Belgian electricity market identifies the drivers and limitations of industrial microgrids in Belgium. The analysis of an industrial electricity bill shows the proportion of the different costs in the total electricity price. Industrial microgrids are usually connected to the MV or HV transmission grid, but the price of transmission only amounts to 7% of the total cost. It is calculated that a decrease in size of a PCC’s transformer (36kV- MT) results in a minor reduction of the connection cost. This suggests that a microgrid must offer additional advantages in order to be economically interesting. In literature, different

103 Chapter 6. Conclusion 104 functional roles for a microgrid can be identiﬁed: they can provide remunerated ancillary services to the overarching utility grid, replace the needs for a UPS installation, accommodate on-site storage and load shifting and allow an integrated optimisation of both the power and heat systems. Several studies indicate that through these means, microgrids can lower the energy cost given the right tariﬀ structures. It is furthermore important that the generating units are optimally sized and dispatched

From an organisational point of view, new business models are necessary to allow the development of microgrids. Four organisational strategies were put forward. Firstly, microgrids developed by the DSO offer advantages because DSOs are one of the industry’s most experi- enced actors regarding development and operation of power systems. However, one problem with DSO operated microgrids is that their role is legally limited to energy distribution. An additional operator would be necessary to manage the production units. A second approach is to create a dedicated entity within the park to implement and operate the microgrid. Such a park-specific cooperative energy supplier allows firms in the park, as well as surrounding households, to invest together in their own, shared electricity supply. A drawback to this strategy is that it can be hard to align a diverse set of stakeholders towards a common goal. A third option is to outsource the operation of the microgrid to an external entity such as an energy service company. Since an ESCO is not bound to a single industrial park, one company could operate several microgrids and specialise in this service. Furthermore, ESCOs offer the advantage that the park’s energy management is coordinated by just one stakeholder. A final suggestion is to allow each firm to invest in their own production. The produced electricity can then be mutually traded between the other firms in the park. Recent developments regarding blockchain technology mean that one to one financial interactions can become quick and reliable and that a highly interactive local power market can be constructed.

Different electrical designs for an industrial microgrid are analysed by optimising the sizing of a grid-connected hybrid power system. A model is developed that is able to simulate the power flows in the microgrid on an hourly basis during its lifetime according to given load profiles, meteorological data and the system parameters. The system parameters consist of the installed solar power, the installed wind power, the installed CHP power, the battery storage capacity, and the maximum battery power.

The model calculates the hourly generated power from the solar and wind units over the course of one year using historical production data. The cogeneration unit follows the electric load according to the instantaneous difference between the renewable production and the park’s electric load profile. The unit has variable thermal and electrical efficiencies depending on its set point. The nominal power-to-heat ratio is 1:1 and there is always maximal heat production. The heat demand that is not produced by the CHP installation is generated using an auxiliary gas boiler. Electrical power can be stored or withdrawn from an electrical Chapter 6. Conclusion 105 storage unit. The storage is controlled using a fuzzy logic based control structure that takes into account the instantaneous electricity price and the state of charge. Any surplus or deficit of electricity is drawn or injected to the utility grid.

The results of the simulation are evaluated according to three evaluation functions: the life cycle cost, the cost of electricity and the CO2 emissions per kWh. Using a genetic optimisation algorithm the system conﬁguration is optimised.

The model is used to analyse several cases with minimal LCC. In the ﬁrst instance a sensitivity analysis is carried out by varying the gas price and the mean electricity selling price. This is done for both current and predicted technology prices. Secondly the impact of an increase in

the ETS CO2 price is assessed. Thirdly a simple demand response program is implemented that flattens the electric load profile. Fourthly, the effect of a gradual ingress of firms in the park, and the subsequent annual increase in load, is analysed. Finally, a multi-objective optimisation is carried out that finds the nondominated set of solutions for the LCC and the

CO2 per kWh.

The results of these simulations suggest the following conclusions.

The optimal system is assessed for a wide range of electricity and gas prices. In all cases there is a high amount of installed wind power. For high gas prices there a tendency towards more solar power and for low gas prices there is a high percentage of power from the CHP installation. The size of the grid connection mainly depends on the installed solar power; the year’s maximum power exchanged with the utility grid is the electricity export during high peaks in solar production on sunny summer days with low demand. In practically all cases there is a netto energy export to the macrogrid which indicates that locally produced power is less costly.

For maximal installed renewable production the renewable power sources are able to supply nearly all the park’s annual energy demand. However, the intermittent behaviour of these sources and the current price structure means that a solid grid connection is generally favourable.

The implemented storage control structure is not able to oﬀer any economic incentives for the deployment of battery storage. Even simulations with a low battery installation cost do not give consistent results. This indicates that new remuneration methods are needed in order to successfully introduce this technology.

The results for the battery storage demonstrate that a genetic algorithm can converge on a local minimum. In order to avoid this behaviour, large sets of simulations were run for varying input parameters. In general, the results show clear and realistic trends which indicates their validity. Chapter 6. Conclusion 106

For future prices of renewable generation there is maximal renewable generation for all of the assessed electricity and gas prices. The results demonstrate a general decrease in the LCC compared to the reference case. This suggests that renewable technologies can lead to a decrease of the energy component in the total electricity bill. However, the subsequent higher production peaks can result in higher distribution costs.

A small amount of demand response can have a large impact on the size of the installation. According to the simulations, a daily load shift of 2.5% on energy basis will result in an optimal system that can be 15% smaller in size. However, the modelled demand response program and used price structure results in a higher cost per kWh.

A gradual ingress of firms in the park and the subsequent sloped annual energy demand have a significant effect on the power system’s optimal configuration. A low energetic demand during the first ten years of the park limits the profitability of on-site renewable infrastructure. Cogeneration is no longer part of the optimal system configuration, and the solar fraction is reduced. Instead, the optimal system has a strong grid dependency. The resulting cost of electricity is about one cent per kWh higher compared to the reference case without gradual ingress.

An optimally conﬁgured microgrid results in a reduction of CO2 emissions by half compared to the reference emissions for grid power in Belgium. An increase in the ETS

price according to the EU’s predictions can further lower the relative CO2 emissions but only slightly. It also leads to a higher LCC.

A multi-objective optimisation identiﬁes the speciﬁc cost of decreasing CO2 emissions. A pareto optimal set of solutions is found between a system with an LCC of 189.1 million e

and CO2 emissions of 95.6gCO2/kWh and a system with an LCC of 202.8 million e and

53.9gCO2/kWh. Increasing the solar share results in a ﬁrst decrease in emissions. A further decrease can be achieved by reducing the size of the CHP installation. This is accompanied by an increase of power from the grid.

In future research, many of these conclusions could be studied in greater depth. The model developed in this study can be enhanced in several ways to include additional aspects of microgrid development.

The organisational aspects of a collective electrical energy supply could be included in the optimisation by constructing an interaction model between the involved actors (firms, DSO, park developer, etc.). Instead of optimising the global system cost, this model can allow the economic impact for each the stakeholders to be assessed individually. This will help define a more detailed suggestion of how an optimised configuration can be translated into an actual inter-firm power system. Chapter 6. Conclusion 107

Since the price of distribution plays such an important role in the total cost of electricity, a more detailed analysis of the interaction with the distribution grid could reveal opportunities for cost savings. Furthermore, a centralised microgrid controller that optimises the dispatch of all units simultaneously can lead to better system performance.

Finally, the issue with the sloped annual energetic demand can be addressed by considering the system size as a dynamic variable. On-site generation that grows according to the annual demand can be included in the model by introducing a multidimensional sizing matrix. The technical system compatibility with expansion and the impact on the local grid facilities can be studied in greater detail. Bibliography

[1] H. Zhao, H. Zhao, and S. Guo, “Evaluating the comprehensive beneﬁt of eco-industrial parks by employing multi-criteria decision making approach for circular economy,” Journal of Cleaner Production, vol. 142, pp. 2262–2276, 2017.

[2] United Nations World Commission on Environment and Development, “Our Common Future,” 1987.

[3] M. R. Chertow, “Industrial Symbiosis: Literature and Taxonomy,” Annual Review of Energy and the Environment, vol. 25, no. 1, pp. 313–337, 2000.

[4] IPCC, “Climate Change 2013 - The Physical Science Basis,” Tech. Rep., 2014.

[5] Eurostat, “Energy balance sheets,” pp. 345–52, 2016.

[6] European Environmental Agency, “Final electricity consumption by sector , EU-27,” p. European Environment Agency, 2013.

[7] R. P. Cˆot´eand E. Cohen-Rosenthal, “Designing eco-industrial parks: a synthesis of some experiences,” Journal of Cleaner Production, vol. 6, no. 3-4, pp. 181–188, 1998.

[8] G. V. Eetvelde, K. Deridder, S. Segers, T. Maes, and M. Crivits, “Sustainability scan- ning of eco-industrial parks,” in 11th European Roundtable on Sustainable Consumption and Production, 2007, pp. 1–20.

[9] E. Kuznetsova, E. Zio, and R. Farel, “A methodological framework for Eco-Industrial Park design and optimization,” Journal of Cleaner Production, vol. 126, pp. 308–324, 2015.

[10] R. A. Frosch and N. E. Gallopoulos, “Strategies for Manufacturing,” Scientiﬁc Amer- ican, vol. 261, no. 3, pp. 144–152, 1989.

[11] J. Timmerman, C. Deckmyn, L. Vandevelde, and G. Van Eetvelde, Low carbon business park manual : a guide for developing and managing energy eﬃcient and low carbon businesses and business parks, 2014.

[12] R. R. Heeres, W. J. V. Vermeulen, and F. B. De Walle, “Eco-industrial park initiatives in the USA and the Netherlands: First lessons,” Journal of Cleaner Production, vol. 12, no. 8-10, pp. 985–995, 2004.

[13] T. Tudor, E. Adam, and M. Bates, “Drivers and limitations for the successful development and functioning of EIPs (eco-industrial parks): A literature review,” Ecological Economics, vol. 61, no. 2-3, pp. 199–207, 2007.

108 Bibliography 109

[14] D. Gibbs and P. Deutz, “Reﬂections on implementing industrial ecology through eco- industrial park development,” Journal of Cleaner Production, vol. 15, no. 17, pp. 1683– 1695, 2007.

[15] R. Dekker, “Dekker, R. (1997). Duurzame ontwikkeling van bedrijventerreinen. ROM Magazine 7/8, 16-18 No Title,” ROM Magazine 7/8, vol. 16-18, 1997.

[16] J. Ehrenfeld and N. Gertler, “Industrial Ecology in Practice: The Evolution of Inter- dependence at Kalundborg,” Journal of Industrial Ecology, vol. 1, no. 1, pp. 67–79, 1997.

[17] DONG Energy, “One step closer to a green future for Asnæs Power Station,” 2015. [Online]. Available: http://www.dongenergy.com/en/media/newsroom/news/articles/ one-step-closer-to-a-green-future-for-asn OT1 ae s-power-station {\ \ } [18] W. Fichtner, M. Frank, and O. Rentz, “Inter-ﬁrm energy supply concepts: An option for cleaner energy production,” Journal of Cleaner Production, vol. 12, no. 8-10, pp. 891–899, 2004.

[19] B. Zwaenepoel, J. Vansteenbrugge, T. L. Vandoorn, G. Van Eetvelde, and L. Vandevelde, “Renewable energy balancing with thermal grid support,” Chemical Engineering Transactions, vol. 35, pp. 535–540, 2013.

[20] R. Lasseter, “MicroGrids,” in IEEE Power Engineering Society Winter Meeting. Con- ference Proceedings, vol. 1, 2002, pp. 305–308.

[21] K. V. Hende and C. Wouters, “The Regulation of Microgrids in Liberalized Electricity Markets in the EU and East Asia,” European Networks Law and Regulation Quarterly, vol. 3, pp. 190–204, 2014.

[22] R. Hanna, M. Ghonima, J. Kleissl, G. Tynan, and D. G. Victor, “Evaluating business models for microgrids: Interactions of technology and policy,” Energy Policy, vol. 103, no. August 2016, pp. 47–61, 2017.

[23] M. C. Lott and S.-I. Kim, “Technology Roadmap Energy storage,” p. 64, 2014.

[24] P. Reyniers, “De reis naar een ’groen’ energielandschap,” 2015.

[25] Eurostat, “Supply, transformation and consumption of electricity - annual data,” 2014. [Online]. Available: http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset= nrg 105a&lang=en

[26] P. Nema, R. K. Nema, and S. Rangnekar, “A current and future state of art development of hybrid energy system using wind and PV-solar: A review,” Renewable and Sustainable Energy Reviews, vol. 13, no. 8, pp. 2096–2103, 2009. Bibliography 110

[27] Chartered Institution of Building Services Engineers, “Introducing CHP,” pp. 1–3, 2012.

[28] Federaal Planbureau and FOD Economie, “Studie over de perspectieven van elektri- citeitsbevoorrading 2030,” p. 244, 2015.

[29] G. Fuchs, B. Lunz, M. Leuthold, and D. U. Sauer, “Technology Overview on Electricity Storage,” p. 66, 2012.

[30] M. H. Shoreh, P. Siano, M. Shaﬁe-khah, V. Loia, and J. P. S. Catal??o, “A survey of industrial applications of Demand Response,” Electric Power Systems Research, vol. 141, pp. 31–49, 2016.

[31] F. Shariatzadeh, P. Mandal, and A. K. Srivastava, “Demand response for sustainable energy systems: A review, application and implementation strategy,” Renewable and Sustainable Energy Reviews, vol. 45, pp. 343–350, 2015.

[32] N. Zhou, N. Liu, J. Zhang, and J. Lei, “Multi-objective optimal sizing for battery storage of PV-based microgrid with demand response,” Energies, vol. 9, no. 8, pp. 1–24, 2016.

[33] B. Zhao, X. Zhang, P. Li, K. Wang, M. Xue, and C. Wang, “Optimal sizing, operating strategy and operational experience of a stand-alone microgrid on Dongfushan Island,” Applied Energy, vol. 113, pp. 1656–1666, 2014.

[34] T. Ma, H. Yang, L. Lu, and J. Peng, “Technical feasibility study on a standalone hybrid solar-wind system with pumped hydro storage for a remote island in Hong Kong,” Renewable Energy, vol. 69, pp. 7–15, 2014.

[35] Stad Gent, “Gents Klimaatplan 2014-2019,” 2014.

[36] Bouwlogie, “Trias Energetica,” 2014. [Online]. Available: http://www.bouwlogie.nl/ trias-energetica/

[37] European Commission, “Liberalisation of the electricity and gas markets,” 2012. [Online]. Available: http://ec.europa.eu/competition/sectors/energy/overview en.html

[38] M. van Werven and M. Scheepers, “DISPOWER The Changing Role of Energy Suppliers and Distribution System Operators in the Deployment of Distributed Generation in Liberalised Electricity Markets,” p. 48, 2005.

[39] J. Kreusel, “Entering a new epoch A brief history of the electric power supply,” ABB review, no. 4, pp. 46–53, 2014.

[40] Next Kraftwerke, “Power and Market,” 2017. [Online]. Available: https: //www.next-kraftwerke.be/en/knowledge-hub/players-in-the-belgian-power-market/ Bibliography 111

[41] Clean Energy Wire, “Electricity market,” 2017. [Online]. Available: https: //www.cleanenergywire.org/tags/electricity-market

[42] VREG, “Energy Market.” [Online]. Available: http://www.vreg.be/en/energy-market

[43] “Elia : facilitator in the free electricity market,” Tech. Rep.

[44] B. Scott, J. P. Chaves-Avila,´ C. Batlle, and I. J. P´erez-Arriaga,“The Value of Aggreg- ators in Electricity Systems,” 2016.

[45] VREG, “Naar wie gaan mijn centen? Energiefactuur bedrijf,” 2016. [On- line]. Available: http://www.vreg.be/sites/default/ﬁles/uploads/naar wie gaan mijn centen energiefactuur bedrijf.pdf

[46] ELIA, “Connection tariﬀs 2016-2019,” 2016.

[47] PricewaterhouseCoopers, “A European comparison of electricity and gas prices for large industrial consumers,” 2015.

[48] VREG, “V-Test,” 2017. [Online]. Available: https://vtest.vreg.be/

[49] ——, “Energiewetgeving.” [Online]. Available: http://www.vreg.be/nl/ energiewetgeving

[50] Vlaamse Overheid, “Energiebesparen voor bedrijven.” [Online]. Available: https: //www.energiesparen.be/subsidies/bedrijven

[51] European Commission, “EU ETS: Development in phases,” 2005. [Online]. Available: http://ec.europa.eu/clima/policies/ets/index en.htm

[52] M. Muˆuls,J. Colmer, R. Martin, and U. J. Wagner, “Evaluating the EU Emissions Trading System : Take it or leave it ? An assessment of the data after ten years,” 2016.

[53] European Commission, “EU Reference Scenario 2016: Energy, Transport and GHG emissions Trends to 2050,” 2016. [Online]. Available: https://ec.europa.eu/energy/en/ data-analysis/energy-modelling

[54] IEA, “Distributed Generation in Liberalised Electricity Markets,” International symposium on distributed generation power system and market aspects, 2002.

[55] G. Pepermans, J. Driesen, D. Haeseldonckx, R. Belmans, and W. D’haeseleer, “Dis- tributed generation: Deﬁnition, beneﬁts and issues,” Energy Policy, vol. 33, no. 6, pp. 787–798, 2005.

[56] R. L. Dohn, “The business case for microgrids,” Siemens AG, pp. 1–9, 2011. Bibliography 112

[57] T. L. Vandoorn and L. Vandevelde, “Microgrids to the the Smart Grid,” in Smart grids : clouds, communications, open source, and automation, D. Bakken and K. Iniewski, Eds. CRC Press, 2012, ch. 9, pp. 191–211.

[58] P. Sreedharan, J. Farbes, E. Cutter, C. K. Woo, and J. Wang, “Microgrid and renewable generation integration: University of California, San Diego,” Applied Energy, vol. 169, pp. 709–720, 2016.

[59] E. Park and S. J. Kwon, “Solutions for optimizing renewable power generation systems at Kyung-Hee University’s Global Campus, South Korea,” Renewable and Sustainable Energy Reviews, vol. 58, pp. 439–449, 2016.

[60] S. E. Jørgensen and S. N. Nielsen, “A carbon cycling model developed for the renewable Energy Danish Island, Samsø,” Ecological Modelling, vol. 306, pp. 106–120, 2015.

[61] A. Majzoobi and A. Khodaei, “Application of microgrids in providing ancillary services to the utility grid,” Energy, vol. 123, pp. 555–563, 2017.

[62] J. F. Sanz, G. Matute, H. Bludszuweit, and E. Laporta, “Microgrids , a new business model for the energy market,” in International Conference on Renewewable Energies and Power Quality, no. 12, 2014.

[63] P. Asmus and M. Lawrence, “Emerging Microgrid Business Models,” Navigant, p. 14, 2016.

[64] M. Soshinskaya, W. H. J. Crijns-Graus, J. M. Guerrero, and J. C. Vasquez, “Microgrids: Experiences, barriers and success factors,” Renewable and Sustainable Energy Reviews, vol. 40, pp. 659–672, 2014.

[65] E. Vine, “An international survey of the energy service company ESCO industry,” Energy Policy, vol. 33, no. 5, pp. 691–704, 2005.

[66] J. Green and P. Newman, “Citizen utilities: The emerging power paradigm,” Energy Policy, vol. 105, no. June 2016, pp. 283–293, 2017.

[67] J. J. Sikorski, J. Haughton, and M. Kraft, “Blockchain technology in the chemical industry: Machine-to-machine electricity market,” Applied Energy, vol. 195, no. 178, pp. 234–246, 2017.

[68] M. E. Peck, “A Microgrid Grows in Brooklyn - Scientific American,” Scientific American, 2016. [Online]. Available: https://www.scientificamerican.com/article/ a-microgrid-grows-in-brooklyn/ Bibliography 113

[69] C. Stevanoni, Z. D. Gr, O. Deblecker, and P. Couneson, “Long-Term Planning of In- dustrial Microgrids,” in 8th IEEE Young Researchers Symposium in Electrical Power Engineering, 2016.

[70] G. Celli, F. Pilo, G. Pisano, and G. Soma, “Optimal participation of a microgrid to the energy market with an intelligent EMS,” 2005 International Power Engineering Conference, pp. 663–668 Vol. 2, 2005.

[71] E. Koutroulis, D. Kolokotsa, A. Potirakis, and K. Kalaitzakis, “Methodology for optimal sizing of stand-alone photovoltaic/wind-generator systems using genetic algorithms,” Solar Energy, vol. 80, no. 9, pp. 1072–1088, 2006.

[72] Y. Li, X. Chen, J. Xiao, and X. Wei, “Optimal conﬁguration for distributed generations in micro-grid system considering diesel as the main control source,” in International Conference on Advanced Mechatronic Systems, 2014, pp. 552–556.

[73] S. M. M. Tafreshi, H. A. Zamani, S. M. Ezzati, M. Baghdadi, and H. Vahedi, “Op- timal unit sizing of distributed energy resources in microgrid using genetic algorithm,” Proceedings - 2010 18th Iranian Conference on Electrical Engineering, ICEE 2010, pp. 836–841, 2010.

[74] J. P. Fossati, A. Galarza, A. Martin-Villate, and L. Fontan, “A method for optimal sizing energy storage systems for microgrids,” Renewable Energy, vol. 77, pp. 539–549, 2015.

[75] M. S. Ismail, M. Moghavvemi, and T. M. I. Mahlia, “Genetic algorithm based optimization on modeling and design of hybrid renewable energy systems,” Energy Conversion and Management, vol. 85, pp. 120–130, 2014.

[76] H. Yang, Z. Wei, and L. Chengzhi, “Optimal design and techno-economic analysis of a hybrid solar-wind power generation system,” Applied Energy, vol. 86, no. 2, pp. 163–169, 2009.

[77] D. Abbes, A. Martinez, and G. Champenois, “Life cycle cost, embodied energy and loss of power supply probability for the optimal design of hybrid power systems,” Mathem- atics and Computers in Simulation, vol. 98, pp. 46–62, 2014.

[78] R. Dufo-L´opez and J. L. Bernal-Agust´ın, “Multi-objective design of PV-wind-diesel- hydrogen-battery systems,” Renewable Energy, vol. 33, no. 12, pp. 2559–2572, 2008.

[79] Y. Katsigiannis, P. Georgilakis, and E. Karapidakis, “Multiobjective genetic algorithm solution to the optimum economic and environmental performance problem of small autonomous hybrid power systems with renewables,” IET Renewable Power Generation, vol. 4, no. 5, p. 404, 2010. Bibliography 114

[80] S. R. Tito, T. T. Lie, and T. N. Anderson, “Optimal sizing of a wind-photovoltaic- battery hybrid renewable energy system considering socio-demographic factors,” Solar Energy, vol. 136, pp. 525–532, 2016.

[81] A. Gonz´alez,J. R. Riba, A. Rius, and R. Puig, “Optimal sizing of a hybrid grid- connected photovoltaic and wind power system,” Applied Energy, vol. 154, pp. 752–762, 2015.

[82] A.´ A. Bayod-R´ujula,M. E. Haro-Larrod´e,and A. Mart´ınez-Gracia,“Sizing criteria of hybrid photovoltaic-wind systems with battery storage and self-consumption considering interaction with the grid,” Solar Energy, vol. 98, no. PC, pp. 582–591, 2013.

[83] W. L. Theo, J. S. Lim, S. R. Wan Alwi, N. E. Mohammad Rozali, W. S. Ho, and Z. Abdul-Manan, “An MILP model for cost-optimal planning of an on-grid hybrid power system for an eco-industrial park,” Energy, vol. 116, pp. 1423–1441, 2016.

[84] A. Kamjoo, A. Maheri, G. A. Putrus, and A. M. Dizqah, “Optimal sizing of grid- connected hybrid wind-PV systems with battery bank storage,” in World Renewable Energy Forum, 2012.

[85] J. Li, W. Wei, and J. Xiang, “A simple sizing algorithm for stand-alone PV/Wind/Bat- tery hybrid microgrids,” Energies, vol. 5, no. 12, pp. 5307–5323, 2012.

[86] H. Yang, L. Lu, and W. Zhou, “A novel optimization sizing model for hybrid solar-wind power generation system,” Solar Energy, vol. 81, no. 1, pp. 76–84, 2007.

[87] I. Gonz´alez-Aparicio and A. Zucker, “Impact of wind power uncertainty forecasting on the market integration of wind energy in Spain,” Applied Energy, vol. 159, pp. 334–349, 2015.

[88] I. Gonz´alez-Aparicio,A. Zucker, F. Careri, F. Monforti, T. Huld, and J. Badger, “EM- HIRES dataset Part I : Wind power generation,” 2016.

[89] G. Schubert, “Modeling hourly electricity generation from PV and wind plants in Europe,” 9th International Conference on the European Energy Market, EEM 12, pp. 1–7, 2012.

[90] E. L. Petersen, N. G. Mortensen, L. Landberg, J. Højstrup, and H. P. Frank, “Wind power meteorology. Part I: climate and turbulence,” Wind Energy, vol. 1, no. 1, pp. 2–22, 1998.

[91] P. Ghadimi, S. Kara, and B. Kornfeld, “The optimal selection of on-site CHP systems through integrated sizing and operational strategy,” Applied Energy, vol. 126, pp. 38–46, 2014. Bibliography 115

[92] ONSITE SYCOM Energy Corporation, “Review of Combined Heat and Technologies Power Technologies,” U.S. Department of Energy, Oﬃce of Energy Eﬃciency and Re- newable Energy, Tech. Rep., 1999.

[93] “CHP.” [Online]. Available: http://www.dynamicenergyusa.com/solutions/ combined-heat-power/

[94] N. Haga, V. Kortela, and A. Ahnger, “Smart Power Generation - District heating solutions,” p. 24, 2012.

[95] A. Colmenar-Santos, E. Rosales-Asensio, D. Borge-Diez, and F. Mur-P´erez,“Cogen- eration and district heating networks: Measures to remove institutional and ﬁnancial barriers that restrict their joint use in the EU-28,” Energy, vol. 85, pp. 403–414, 2017.

[96] K. Darrow, R. Tidball, J. Wang, and A. Hampson, “Catolog of CHP Technologies,” U.S. Environmental Protection Agency, Tech. Rep. March, 2015.

[97] Z. Beihong and L. Weiding, “An optimal sizing method for cogeneration plants,” Energy and Buildings, vol. 38, no. 3, pp. 189–195, 2006.

[98] S. Gamou, R. Yokoyama, and K. Ito, “Optimal unit sizing of cogeneration systems in consideration of uncertain energy demands as continuous random variables,” Energy Conversion and Management, vol. 43, no. 9-12, pp. 1349–1361, 2002.

[99] H. Ren, W. Gao, and Y. Ruan, “Optimal sizing for residential CHP system,” Applied Thermal Engineering, vol. 28, no. 5-6, pp. 514–523, 2008.

[100] D. Zhang, S. Evangelisti, P. Lettieri, and L. G. Papageorgiou, “Optimal design of CHP-based microgrids: Multiobjective optimisation and life cycle assessment,” Energy, vol. 85, pp. 181–193, 2015.

[101] H. Wang, W. Yin, E. Abdollahi, R. Lahdelma, and W. Jiao, “Modelling and optimization of CHP based district heating system with renewable energy production and energy storage,” Applied Energy, vol. 159, pp. 401–421, 2015.

[102] M. Nazari-Heris, S. Abapour, and B. Mohammadi-Ivatloo, “Optimal economic dispatch of FC-CHP based heat and power micro-grids,” Applied Thermal Engineering, vol. 114, pp. 756–769, 2017.

[103] R. Lahdelma and H. Hakonen, “An eﬃcient linear programming algorithm for combined heat and power production,” European Journal of Operational Research, vol. 148, no. 1, pp. 141–151, 2003. Bibliography 116

[104] M. Alipour, K. Zare, and B. Mohammadi-Ivatloo, “Optimal risk-constrained participation of industrial cogeneration systems in the day-ahead energy markets,” Renewable and Sustainable Energy Reviews, vol. 60, pp. 421–432, 2016.

[105] C. Milan, M. Stadler, G. Cardoso, and S. Mashayekh, “Modeling of non-linear CHP eﬃciency curves in distributed energy systems,” Applied Energy, vol. 148, pp. 334–347, 2015.

[106] L. Mongibello, N. Bianco, M. Caliano, and G. Graditi, “Comparison between two different operation strategies for a heat-driven residential natural gas-ﬁred CHP system: Heat dumping vs. load partialization,” Applied Energy, vol. 184, pp. 55–67, 2016.

[107] X. Luo, J. Wang, M. Dooner, and J. Clarke, “Overview of current development in electrical energy storage technologies and the application potential in power system operation,” Applied Energy, vol. 137, pp. 511–536, 2015.

[108] X. Tan, Q. Li, and H. Wang, “Advances and trends of energy storage technology in Microgrid,” International Journal of Electrical Power and Energy Systems, vol. 44, no. 1, pp. 179–191, 2013.

[109] European Commission, “Commercialisation of Energy Storage in Europe,” 2015.

[110] H. Yang, W. Zhou, L. Lu, and Z. Fang, “Optimal sizing method for stand-alone hybrid solar-wind system with LPSP technology by using genetic algorithm,” Solar Energy, vol. 82, no. 4, pp. 354–367, 2008.

[111] A. Chaouachi, R. M. Kamel, R. Andoulsi, and K. Nagasaka, “Multiobjective Intelligent Energy Management for a Microgrid,” IEEE Transactions on Industrial Electronics, vol. 60, no. 4, pp. 1688–1699, 2013.

[112] J. P. Fossati, A. Galarza, A. Martin-Villate, J. M. Echeverria, and L. Fontan, “Optimal scheduling of a microgrid with a fuzzy logic controlled storage system,” International Journal of Electrical Power and Energy Systems, vol. 68, pp. 61–70, 2015.

[113] Mathworks, Fuzzy Logic Toolbox User’s Guide, 2017.

[114] F. Dernoncourt, “Introduction to Fuzzy Logic Control,” 2010.

[115] P. Palensky and D. Dietrich, “Demand side management: Demand response, intelligent energy systems, and smart loads,” IEEE Transactions on Industrial Informatics, vol. 7, no. 3, pp. 381–388, 2011.

[116] P. S. Moura and A. T. de Almeida, “Multi-objective optimization of a mixed renewable system with demand-side management,” Renewable and Sustainable Energy Reviews, vol. 14, no. 5, pp. 1461–1468, 2010. Bibliography 117

[117] M. Starke, N. Alkadi, and O. Ma, Assessment of Industrial Load for Demand Response across U . S . Regions of the Western Interconnect, 2013, no. September.

[118] B. Kirby, “Spinning Reserve From Responsive Loads,” 2003.

[119] H. C. Gils, “Assessment of the theoretical demand response potential in Europe,” En- ergy, vol. 67, pp. 1–18, 2014.

[120] S. Schlomer, T. Bruckner, L. Fulton, E. Hertwich, A. McKinnon, D. Perczyk, J. Roy, R. Schaeﬀer, R. Sims, P. Smith, and R. Wiser, “Annex III: Technology-speciﬁc cost and performance parameters,” Climate Change 2014: Mitigation of Climate Change. Con- tribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, pp. 1329–1356, 2014.

[121] M. Messagie, J. Mertens, L. Oliveira, S. Rangaraju, J. Sanfelix, T. Coosemans, J. Van Mierlo, and C. Macharis, “The hourly life cycle carbon footprint of electricity generation in Belgium, bringing a temporal resolution in life cycle assessment,” Applied Energy, vol. 134, no. August, pp. 469–476, 2014.

[122] International Energy Agency, Nuclear Energy Agency, and Organisation for Economic Co-operation and Development, “Projected Costs of Generating Electricity 2015,” Tech. Rep., 2015.

[123] European Commission, “Energy Roadmap 2050,” 2012.

[124] F. Millo, L. Rolando, R. Fuso, and F. Mallamo, “Real CO2 emissions beneﬁts and end user’s operating costs of a plug-in Hybrid Electric Vehicle,” Applied Energy, vol. 114, pp. 563–571, 2014. 118 119 Numerical simulation results

Table 1: Sensitivity analysis for current market prices

Price of gas (ce/kWh) 3.28 4.28 5.28 6.28 7.28 8.28 9.28 10.28 11.28 12.28 LCC 1.40e8 1.57e8 1.72e8 1.86e8 1.99e8 2.12e8 2.24e8 2.35e8 2.45e8 2.54e8 10.47 COE* 0.0778 0.085 0.090 0.095 0.099 0.103 0.106 0.109 0.111 0.113 CO2 260.37 137.33 111.05 105.63 93.37 84.02 82.42 68.82 42.68 41.11 LCC 1.41e8 1.59e8 1.74e8 1.88e8 2.02e8 2.15e8 2.27e8 2.39e8 2.50e8 2.61e8 11.47 COE* 0.0785 0.0861 0.0914 0.0961 0.1011 0.1048 0.1082 0.1116 0.1143 0.1168 CO2 260.06 119.93 116.56 100.24 95.76 91.91 82.30 74.71 66.94 63.86 LCC 1.41e8 1.61e8 1.76e8 1.90e8 2.04e8 2.17e8 2.30e8 2.42e8 2.54e8 2.65e8 12.47 COE* 0.0787 0.0875 0.0928 0.0973 0.1019 0.1063 0.1101 0.1137 0.1169 0.1199 CO2 275.63 130.46 110.75 103.04 97.44 88.10 84.35 81.97 76.72 69.44 LCC 1.43e8 1.63e8 1.78e8 1.91e8 2.06e8 2.19e8 2.32e8 2.45e8 2.57e8 2.69e8 13.47 COE* 0.0800 0.0885 0.0940 0.0985 0.1033 0.1074 0.1116 0.1156 0.1190 0.1223 CO2 258.27 119.75 114.06 101.06 100.94 92.48 86.72 84.29 81.95 75.67 /kWh) e LCC 1.42e8 1.64e8 1.79e8 1.93e8 2.07e8 2.21e8 2.34e8 2.47e8 2.60e8 2.72e8 14.47 COE* 0.0796 0.0897 0.0947 0.0997 0.1045 0.1087 0.1128 0.1170 0.1210 0.1245 CO2 275.52 121.02 113.55 103.72 102.07 91.32 91.27 87.08 84.25 82.58 LCC 1.43e8 1.66e8 1.81e8 1.95e8 2.10e8 2.23e8 2.36e8 2.49e8 2.62e8 2.75e8 15.47 COE* 0.0802 0.0908 0.0959 0.1005 0.1059 0.1101 0.1145 0.1184 0.1225 0.1266 CO2 292.38 116.93 112.38 103.06 101.64 95.96 91.64 88.49 83.91 82.43 LCC 1.43e8 1.68e8 1.82e8 1.96e8 2.10e8 2.10e8 2.38e8 2.52e8 2.66e8 2.77e8

Mean price of electricity (c 16.47 COE* 0.0800 0.0918 0.0967 0.1018 0.1065 0.1109 0.1154 0.1201 0.1248 0.1277 CO2 280.60 117.84 105.60 99.92 99.92 92.59 91.87 92.51 94.14 86.91 LCC 1.43e8 1.69e8 1.84e8 1.98e8 2.12e8 2.26e8 2.40e8 2.54e8 2.66e8 2.80e8 17.47 COE* 0.0803 0.0928 0.0984 0.1026 0.1076 0.1121 0.1172 0.1213 0.1251 0.1296 CO2 280.59 120.26 103.02 100.05 101.18 94.09 95.00 92.98 91.61 90.26 LCC 1.44e8 1.70e8 1.85e8 2.00e8 2.14e8 2.28e8 2.42e8 2.55e8 2.68e8 2.81e8 18.47 COE* 0.0810 0.0936 0.0991 0.1043 0.1086 0.1133 0.1180 0.1222 0.1265 0.1306 CO2 284.64 112.35 107.47 105.96 95.70 93.20 94.06 93.27 91.65 92.76 LCC 1.46e8 1.73e8 1.86e8 2.02e8 2.16e8 2.30e8 2.43e8 2.56e8 2.70e8 2.84e8 19.47 COE* 0.0810 0.0952 0.0996 0.1053 0.1099 0.1150 0.1190 0.1230 0.1273 0.1320 CO2 266.52 112.01 102.85 102.64 99.67 100.46 94.36 93.45 92.25 91.32 Table 2: Sensitivity analysis for future market prices

Price of gas (ce/kWh) 4.28 7.28 11.28 LCC 1.27e8 1.66e8 2.15e8 11.47 COE* 0.0649 0.0771 0.0912

/kWh) CO2 94.28 87.93 69.73 e LCC 1.31e8 1.72e8 2.25e8 14.47 COE* 0.0675 0.0807 0.0977 CO2 94.70 93.15 85.35 LCC 1.36e8 1.78e8 2.32e8 18.47 COE* 0.0711 0.0847 0.1024

Price of electricity (c CO2 95.01 93.66 91.08