THIS PAPER WAS PRESENTED AT THE 4TH SOLAR INTEGRATION WORKSHOP (2014) AND PUBLISHED IN THE WORKSHOP’S PROCEEDINGS. 1 A Proposal for New Requirements for the Fault Behaviour of Connected to Low Voltage Networks Emmanuel van Ruitenbeek, Jens C. Boemer, Member, IEEE, Konstantinos Skaloumpakas, Jose´ Luis Rueda Torres, Senior Member, IEEE, Madeleine Gibescu, Member, IEEE, and Mart A.M.M. van der Meijden, Member, IEEE

Abstract—Under current requirements in Europe drops below 0.8 p.u. [3]. European and North American and North America, small scale, low voltage connected dis- standard also do not require LVRT capability for small scale tributed generators generally disconnect when the voltage at LV connected DG [4], [5]. This creates a situation of potential their point of common coupling drops due to a fault in the transmission system. With an expected increase of low voltage massive disconnection of PV systems. Transmission system connected photovoltaic systems, massive disconnection of these faults and consequent voltage dip propagation throughout the generators may cause post-fault active power balance prob- power system might lead to significant post-fault active power lems. This paper investigates the impact of large amounts of imbalance and thus frequency instability. To prevent problems photovoltaic installations in low voltage networks on post-fault similar to those caused by unfavourable frequency protection active power balance following transmission system faults. A test system is constructed with distributed generation penetration settings [6], [7], the adequacy of current and recently proposed representative of the German network. The test system comprises grid connection requirements (GCR) with respect to post-fault all voltage levels from low to extra high, with the low and active power balance must be investigated. medium voltage levels simplified by means of aggregation. Voltage ENTSO-E has included low voltage ride-through (LVRT) in low voltage busbars and post-fault active power output of capability requirements in the final version of the Network distributed generators are investigated for different operational scenarios and network fault control modes. Based on evaluations Code published in March 2013 [8] for large-scale generators on the representative test system, new requirements for the fault of Type B, which are usually connected to MV networks and behaviour of distributed generation connected to low voltage have a maximum capacity between 1 and 50 MW. However, no networks are proposed. The recommendations are differentiated other GCR in Europe or North America requires LV connected into short- and long-term and consider strategies to keep existing DG to ride through faults. The need for such requirements installations in the low voltage networks online. has, however, been identified worldwide [9], [10], [11]. In- Index Terms—photovoltaic systems; network fault response; depth studies to date have been few. Recently results focusing low voltage networks; grid codes; low voltage ride-through; primarily on system voltage performance have been published dynamic network support. in [12] and [13].

I.INTRODUCTION The focus in this work lies on the impact of large amounts of PV installations in LV networks on post-fault active power HE share of low voltage (LV) connected distributed balance following transmission system faults. The approach is generation (DG) in serving loads has increased in recent T based on representative modeling of real power systems. The years and will keep increasing in the future. In Germany study horizon is 2022. Two research questions are formulated: especially, LV connected photovoltaic (PV) systems are likely to replace generation from large-scale conventional power 1) How much active power infeed from LV connected DG plants as the installed capacity is expected to increase to almost is lost due to a transmission system fault in a worst case 40 GW in the year 2022 [1], [2]. For low DG penetration lev- scenario if current practices are continued? els, their influence on the power system fault response is negli- 2) How should new LV connected DG behave during gible. During times of very high instantaneous DG penetration transmission system faults? however, DG’s network fault behaviour will determine the • To have a positive impact on system stability; power system’s fault response. Current regulations in Germany • To keep existing LV connected DG online. do not require low voltage ride-through (LVRT) capabilities for Section II describes the methodology of the research and LV connected DG (see figure 1). Instead, these are required the test system with which the simulations are performed. In to disconnect within 200 ms if the voltage at their terminals section III the results are presented and analysed. Finally, the E. van Ruitenbeek, J. C. Boemer, K. Skaloumpakas, J. L. Rueda conclusions are drawn in section IV. and M. A.M.M. van der Meijden are with the Electrical Sustainable En- ergy Department, Delft University of Technology, Delft, the Netherlands. [email protected] M. Gibescu is with the Electrical Energy Systems group, Eindhoven University of Technology, Eindhoven, the Netherlands. [email protected] THIS PAPER WAS PRESENTED AT THE 4TH SOLAR INTEGRATION WORKSHOP (2014) AND PUBLISHED IN THE WORKSHOP’S PROCEEDINGS. 2

1998 2001 2005 2008 2009 2011 2012 2013 Typical installation eHV size per voltage VDN Guideline VDN DG data eHV/HV 2004 TC 2007 level and DG technology HV

SDLWindV 2009 Load and Typical penetration Representative MV VDEW BDEW MV-Guideline MV-Guideline population data levels complete grid

LV VDE-AR-N VDEW 4105 LV-Guideline Typical grid ϕ TC SDL cos(ϕ)=1 cos(ϕ)=const. cos( )=f(P) LVRT structure and data variable power LVRT with dynamic LVRT with dynamic fixed power adjustable power with factor with voltage support acc. to voltage support acc. factor factor ±0,95 blocking mode characteristic curve TC 2007 to SDLWindV 2009 Fig. 2: Creation of representative complete grids. Fig. 1: Historical overview of GCR in Germany for network fault response [14] [15]. Construction of Determination of Construction of Grid data representative equivalent aggregated grids complete grids impedances

II.METHODOLOGY No A. General approach

Satisfactory Construction of The research is conducted with a test system based on the behaviour of Yes different voltage Simulation network New England (NE) system. The NE system is extended to aggregated grids? levels include LV, MV and HV subsystems which are built based on current (2012) and forecasted (2022) data for the German Fig. 3: Overview of the method for determining the equivalent network [2], [16]. network models. A comprehensive test system is constructed, from the LV (0.4 kV) level to extra-high voltage (eHV) (380 kV). Mod- elling the system in a new integral way allows for obtaining re- 2013 [2]. The obtained size may not always correspond to sults that depend on interactions between the voltage levels and the most common size in reality, e.g. when the most common consider the response of ‘active’ distribution systems (ADS) at sizes in reality are 5 and 50 kW, the ‘typical’ size might be the outskirts of a large-area voltage dip. Modelling a complete 20 kW. power system would, however, lead to unacceptable computing Then, for each of the three ‘region types’ (urban, suburban burden for running time domain simulations. Therefore, the and rural) as defined by [16], typical DG penetration levels ADS are aggregated on the LV and MV (20 kV) levels. were defined by combining load and population data provided Aggregation there is justified as those voltage levels have a by a German distribution system operator (DSO) using (1) radial structure and single connection point. DG technologies through (3). Equation (2) shows that it was assumed that gen- included in the test system are PV systems, wind turbines (full eration scales with area, while (3) scales load with population. converters and doubly-fed induction generators) and combined This approach does not consider industrial loads that might be heat & power plants (CHP). spread in a different fashion. Load distribution over voltage The problem is analysed in the time domain using positive- levels was obtained from DSO data. With the penetration levels sequence, root mean square simulations for a balanced three- and typical installation sizes fixed, the number of installations phase transmission system fault distant to the analysed system in each network is determined by the load in the network. area. A three-phase fault is chosen for simulating and evaluat- DGcapacityi ing the system behaviour with respect to a wide spread voltage P enetrationi = · 100 (1) P eakload sag. i Multiple scenarios are considered to study the effects of 1 Areai pre-fault operating point, LV DG fault control modes (CM) DGcapacityi = DGcapacityP LZ−3 · (2) and MV DG full dynamic fault support settings. Results are AreaP LZ−3 compared by merit of their post-fault active power balance and voltage quality. P opulation P eakload = P eakload · i (3) i DSO P opulation B. Test system construction DSO Penetration levels obtained are provided in table V in the To create the test system, first DG, load, population and appendix. These can differ from the penetration levels in the grid data are combined. This creates a representative complete actual test system (see table VI), where it should be pointed out grid, which consists of a typical grid structure with typical that significant differences in the ‘all DG’ category penetration penetration for each DG technology (see figure 2). First a level arise from the omission of hydro- and ‘typical’ installation size was determined for PV systems, wind turbines and CHP plants for all voltage levels by taking 1German postal codes have 5 digits, the first three digits indicate ever the median of the installed DG in Germany connected up to smaller geographic regions. THIS PAPER WAS PRESENTED AT THE 4TH SOLAR INTEGRATION WORKSHOP (2014) AND PUBLISHED IN THE WORKSHOP’S PROCEEDINGS. 3

eHV

2x

HV HV HV HV

Z yielding `V_%1`VR Load Z yielding PF100 similarsimilar aRCI NEW `Q` s1I1C:` PF100 behaviourbehaviour GV.:01Q%`- HV HV

aggregated aggregated DG load by technology Load Fig. 4: Distribution network aggregation. LVRT PF100 LVRT PF100 aRCI

in the implemented test system. The suburban CHP network MV lacks PV penetration by design to provide a clearer insight

MV into the CHP behaviour during a fault. Load NEW PF095 NEW LVRT NEW Representative suburban and rural LV and MV distribution CHP / Bio

PF100 PF100 networks were provided by the same German DSO. DG was Wind introduced in these networks and where necessary network Photovoltaic expansion was implemented according to the guidelines in [17] LV to avoid overloads and/or voltage violations. Three differ- Class specifies the fault behviour LV ent distribution networks are considered. Two variants of a Load NEW PFPOW NEW Class suburban network, one with only a high penetration of PV systems installed at the LV network, the other with only a Fig. 5: General setup of the subsystem model including HV, high penetration of CHP installed at the LV network, and a MV and LV equivalents. PF100, PF095 and PFPOW indicate rural network with high amounts of PV and wind generation. DG without low voltage ride-through capability operated at Figure 3 shows that the next step in constructing the test constant 1, 0.95 (inductive) or dynamic power network is the creation of aggregated equivalent grids. These factor respectively. were constructed and validated according to [18] by use of a simple aggregation method (see figure 4). All generation and load in the network were aggregated and placed behind 29 a single impedance. The impedance is set to give similar 27 fault response behaviour at the connection point. A graphical 25 representation of the aggregated network is provided in figure 5. 15 The aggregated MV networks are then connected to a high 18 voltage (HV) sub-transmission ring network (110 kV) [19] with DG installed based on German network data. The HV 04 39 network is not aggregated, as some (rural) networks are 20 connected to the eHV system at two locations. Simulation performance in this study however demanded that each HV network only used one connection point. 08 Finally the HV networks are connected to the eHV network. Since no German eHV model was available a benchmark system was used instead. The IEEE 39-bus, 10-machine New England Test System [20] was used and adapted to operate at Fig. 6: New England test system with connection points of 380 kV and 50 Hz instead of 220 kV and 60 Hz respectively. ADS and fault location. Each distribution network type is connected at three points in the NE system, making for a total of nine ADS in the test system (see figure 6). Connection points are listed in table C. DG and load modelling I. The ten generators in the NE system are all synchronous Existing DG is modelled according to their actual machines, with automatic voltage regulator (AVR) and power performance defined by GCR in place when they went system stabiliser [21]. The governor is not considered due to into operation. This holds for fault response, as well as for the phenomenon and time frame which are of interest. The (dynamic) power factor control. The only exception are the total amount and a high level overview of networks in the test MV connected wind turbines, as 50 percent of these were system is shown in figure 7. retrofitted to give them LVRT capability to prevent massive THIS PAPER WAS PRESENTED AT THE 4TH SOLAR INTEGRATION WORKSHOP (2014) AND PUBLISHED IN THE WORKSHOP’S PROCEEDINGS. 4

TABLE I: Sub-transmission system connection points. the PV system resynchronises immediately; Area Busbars Type 3) Additional reactive current injection (aRCI): the con- verter injects an additional reactive current into the 1 18, 25, 27 Sub CHP 2 04, 08, 39 Sub PV network during the fault in proportion to the voltage 3 15, 20, 29 Rural dip. If necessary, the active current may be curtailed depending on settings; 4) Additional reactive and active current injection (aRACI): New England similar to aRCI, but adds an additional active current 39-bus eHV 10-machine component. Both the active and reactive components are scaled if converter limits are reached.

3 The k-factor of LV connected PV systems (proportional 3 3 suburban suburban HV rural gain for aRCI and aRACI modes) is set to 6 p.u. to increase PV CHP the impact of the network support. All LVRT control modes for the PV systems use a delayed active power recovery 27 27 48 suburban suburban according to GCR, i.e. active power is restored post-fault MV rural PV CHP with a ramp of 20 percent per second of the pre-fault value.

2268 The models used in the study are based on the 2268 8208 suburban suburban LV rural fully rated wind turbine generator and doubly fed induction PV CHP generator templates provided by Powerfactory [24], [25].

Fig. 7: Overview of test system networks. CHP plants are modelled as directly coupled synchronous generators. The dynamic model introduced in [26] is used, 0% 20% 40% 60% 80% 100% MW which employs an AVR (EXAC1A) in the dynamic model of 2022 the generator. Generator parameters were used as much as eHV 71% 4% 6% 6% 13% 58,512 & HV possible from commercially available generators with similar ratings as those of the models [27]. Other parameters were left to their predefined PowerFactory values. MV 20% 2% 29% 13% 36% 58,578 Aggregated load models presented in [28] are used in all LV, MV and HV networks except for the LV suburban LV 51% 9% 40% 37,410 networks, where a static model is employed (cf. table III). The static part of the load is represented in exponential form, cf. (4) and (5). Parameters are provided in table II. Frequency Sum 47% 1% 2% 12% 2% 7% 28% 154,500 dependency of static loads is ignored, the dynamic part of the load represents motor loads with their particular dynamic COS(phi)=1 COS(phi)=const. COS(phi)(P) LVRT TC SDL NEW characteristics. Loads in the eHV system were left as constant Fig. 8: Installed DG capability in the test system, where LVRT, impedance [20]. TC and SDL refer to BM, [14] and [15] respectively.   V eaP  Pexp = P0 · aP (4) V0 disconnection of in case of a transmission system fault [22]. This creates multiple ‘types’ of e.g. MV connected PV systems, each with different settings. The resulting classes   V eaQ  Qexp = Q0 · aQ (5) and their installed capacities in the aggregated networks are V0 presented in figures 5 and 8 respectively.

The PV system model used is based on the model described TABLE II: Installed load types in representative networks. in [13], [23]. To simplify this model the DC chopper has been (a) Suburban (b) Rural removed by setting a constant voltage for the DC link. The LV connected PV systems labelled as ‘NEW’ (installed between Type [-] Dynamic Type [-] Dynamic 2012 and 2022) can be set to four control modes: load [%] load [%] LV Residential 0 LV Commercial 20 1) No low voltage ride-through (nLVRT): status quo, PV MV Commercial/ 20 MV Commercial/ 20 system is disconnected 100 ms after voltage drops below industrial industrial 0.8 p.u.; HV Mixed 20 HV Mixed 20 2) Blocking mode (BM): id and iq are driven to zero when the voltage drops below 0.8 p.u. After fault clearance THIS PAPER WAS PRESENTED AT THE 4TH SOLAR INTEGRATION WORKSHOP (2014) AND PUBLISHED IN THE WORKSHOP’S PROCEEDINGS. 5

TABLE III: Exponential load parameters [13]. This way, the effects of the different control settings in HV Loads MV loads LV loads 2022 could be compared for three typical scenarios with Mixed Commercial/ Residential Commercial very different instantaneous penetration levels. The 2012 cases industrial serve as a base and show the ‘current’ state of the system. The

aP 1.0 1.0 1.0 1.0 dispatch settings for the operational scenarios are presented in eaP 1.0 1.4 1.7 1.4 figure 10. In total 27 study cases were investigated. However, a 1.0 1.0 1.0 1.0 Q due to restrictions in space, only 2022 OS2 and OS3 are treated eaQ 3.2 5.5 4.7 5.5 here.

Network III.RESULTS AND ANALYSIS All simulations were run for 10 s which was deemed 2012 2022 sufficient for the transient stability time frame. Results are presented for the first 5 s for sake of readability. No further OS 1 OS 2 OS 3 OS 1 OS 2 OS 3 „warm „windy, „windy, „warm „windy, „windy, transients occur after 5 s. The most important results for oper- afternoon“ cloudy“ sunny“ afternoon“ cloudy“ sunny“ ational scenarios 2 and 3 in 2022 are discussed here, as these New LV connected PV fault control modes represent moderate and very high instantaneous penetration of CM 1 CM 2 CM 3 CM 4 DG. nLVRT BM aRCI aRACI Table IV shows that in OS2, wind turbine and CHP output are high, while PV system output is low. Load consumption Full dynamic voltage support in MV networks is also high. Large reverse power flows occur in OS3, with Base Case Alternative Case only new DG at HV/MV-substation all new DG in MV networks all DG output set high while load is low. Voltage profiles are shown for the LV busbar connected to HV busbar number 3, Fig. 9: Overview of study cases. OS is operational scenario. connected to eHV busbar 20 (see figures 5 and 6). At this nLVRT, BM, aRCI and aRACI denote no low voltage ride- busbar, different control modes lead to voltage dips either through, blocking mode, additional reactive current injection below or above 0.8 p.u., illustrating that they can make the and additional reactive and active current injection respec- difference between DG disconnecting or staying online. tively. TABLE IV: Multiplication factors for rated capacity (including 7,000 100 coincidence factors) for operational scenarios 2 and 3. 6,034 6,068 6,077 6,157 90 6,000 81 PV Wind CHP Loadsuburban Loadrural 6,021 6,030 6,041 6,123 80 5,000 OS2 0.40 0.90 0.85 0.90 1.00 70 OS3 0.80 0.90 0.85 0.45 0.50 56 60 4,000 3,649 3,249 50 3,605 3,000 3,222 39 Percent % Percent Megawatt 40 2,973 A. 2022 OS2 2,000 25 24 2,388 30 15 1,834 20 In OS2, the PV systems are set to a low power output and 1,000 1,490 1,480 therefore any PV systems with dynamic power factor control 919 10 0 0 do not exchange any reactive power with the network. OS1 OS2 OS3 OS1 OS2 OS3 2012 2022 Figure 11a shows the voltage in the selected LV busbar Gentotal [MW] Load [MW] GenDG [MW] Share of GenDG [%] for the four control modes of the new LV connected PV systems. Pre-fault voltage is 1.00 p.u. The fault occurs at Fig. 10: Dispatch settings for study cases (see figure 9). 0.1 s and is cleared after 150 ms. The voltage drop during the fault is virtually identical for the nLVRT and BM modes. During the first 100 ms of the fault, the voltage drops just D. Study cases below 0.80 p.u. After 100 ms, the nLVRT LV connected DGs An overview of the study cases is provided in figure 9. Two disconnect and a further voltage drop occurs, down to 0.79 p.u. study years, three operational scenarios (OS) and, for 2022, This sudden voltage drop is caused by the lack of active power four LV connected PV fault control modes and two k-factors of injection in the LV network upon disconnection of DG. To MV connected DG are considered. MV connected PV systems maintain the power balance, power has to be imported from use a k-factor of 0.2 p.u. In practice only HV/MV substation the MV network, causing a voltage drop over the impedance connected PV systems provide full dynamic support during a between the networks. The aRCI and aRACI modes deliver fault with a k-factor of 2 p.u. These account for approximately full dynamic network support by injecting additional current 10 percent of installed MV PV capacity. Simulations are during the fault. The purely reactive power injection by the performed with a k-factor of 0.2 p.u. and 2 p.u. to consider aRCI mode results in a slightly higher voltage during the the consequences of operating all MV connected PV systems fault. Since the LV network has a high R/X ratio this is an with full dynamic support. unexpected result and requires further research. After 100 ms, 1,20 1000, 1000, 300,

DIgSILENT 1,00 800, 800, 200,

0,80 600, 600, 100,

0,60 400, 400, 0,00

0,40 200, 200, -100,

0,20 0,00 0,00 -200, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [-] 5, 0 1, 2, 3, 4, [-] 5,

1,20 1000, 80,0 300,

1,00 800, 64,0 200,

0,80 600, 48,0 100,

0,60 400, 32,0 0,00

0,40THIS PAPER WAS PRESENTED AT THE 4TH200, SOLAR INTEGRATION WORKSHOP (2014)16,0 AND PUBLISHED IN THE WORKSHOP’S-100, PROCEEDINGS. 6 2022 OS2 LV connected DG power 20220,20 OS2 Voltage of eHV-19-2001_HV-03_MV-02_LV-02 0,00 0,00 -200, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [-] 5, 0 1, 2, 3, 4, [-] 5, B. 2022 OS3 1,201,20 1000,1000, 80,01000, 90,0 300, [p.u.] [MW] In OS3, PV system output is high, meaning all installations DIgSILENT 1,001,00 800,800, 64,0800, 60,0 200, with dynamic power factor control are operated with an 0,80 600, 600, 100, 0,80 600, inductive48,0 power factor and thus30,0 consume reactive power 0,600,60 400,400, pre-fault.32,0400, 0,00 0,00 0,400,40 200,200, 16,0200, -30,0 -100,

0,200,20 0,000,00 0,00Figure0,00 12a shows the voltage-60,0 in the-200, selected LV busbar. 00 1,1, 2,2, 3,3, 4, 4,[s][s]5, 5, 0 0 1, 1, 2, 2, 3, 3, 4, [s]4, [s]5, 5, The0 lowest0 1, 1, voltage2, 2, 3, reached3, 4, [-]4, 5,[-] in5, the nLVRT0 01, case1,2, is2,3, 0.783,4, p.u.[-] 4,5, [-] 5, (a) Voltage in busbar eHV-20 HV- (b) Active power output by LV DG. 300,031,20MV-02 LV-02. nLVRT in black, 1000,nLVRT1000, in black, BM in red, aRCI in The300,80,0 voltage rises at disconnection300, of300, nLVRT DG due to the BM in red, aRCI in blue, aRACI in blue, aRACI in green. pre-fault reactive power consumption of the DG. For the same 200,1,00 800,800, 200,64,0 200, 200, green. reason, in OS3 the BM actually ‘indirectly’ boosts the voltage 100,0,80 600,600, 100,48,0 100, 100, Fig. 11: 2022 OS2 results during the entirety of the fault, keeping the voltage above 0,000,60 400,400, 0.800,0032,0 p.u. As a result, fewer old nLVRT0,00 0,00 PV systems disconnect. -100,0,40 200,200, The-100,16,0 aRCI and aRACI modes again-100, deliver-100, maximum voltage -200, 0,00 boosting-200, of 0.04 p.u. -200, the0,20 voltage with both modes is 0.830,00 p.u. The voltage then 0,00 -200, 00 1,1, 2,2, 3,3, 4, 4,[-][s]5, 5, 0 0 1, 1, 2, 2, 3, 3, 4, 4,[-] [s]5, 5, The0 0 temporary1, 1, 2, 2, 3, post-fault3, 4, [-]4, 5,[-] overvoltage5, 0 01, for1,2, all2,3, modes3,4, [-] is4,5, [-] 5, drops a further 0.01 p.u. for the aRACI mode. For the aRCI greater than in OS2 at 1.08 p.u. for the nLVRT and BM mode1,20 however, the voltage remains1000, at 0.83 p.u. In this case the 80,0 90,0 modes and 1.07 p.u. for the aRCI and aRACI modes. This second voltage drop is caused by disconnection of old nLVRT 1,00 800, is caused64,0 by the combination of large60,0 reverse power flows LV connected DG in other LV networks, causing larger power 0,80 600, and48,0 the tap-changer settings30,0 [29] as well as flows in the higher voltage levels. 0,60 400, pre-fault32,0 reactive power consumption.0,00 Post-fault steady state Post-fault, the voltage drops to 0.98 and 0.99 p.u. for the undervoltage for the nLVRT mode is larger than in OS2 at nLVRT0,40 and BM cases respectively.200, The voltage returns to 16,0 -30,0 0.05 p.u. The LVRT modes manage to restore the voltage to 1.000,20 p.u. for the aRCI and aRACI0,00 modes. The differences are 0,00 -60,0 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5, 1.00 p.u.0 Again,1, 2, the post-fault3, 4, [-] 5, negative slope0 1, for the2, voltage3, 4, [-] 5, caused by the amount of disconnecting DGs in the system. in the nLVRT case is caused by the lack of governors in the The300, temporary overvoltage occuring1000, immediately after fault 300, 300, central power plants and the related frequency drop. clearance and subsequent undervoltage in the system is 200, 800, Figure200, 12b shows that the pre-fault200, active power output caused by the behaviour of the large synchronous generators 100, 600, by LV100, connected DG is 983 MW.100, In OS3, the remaining and their power system stabilisers in the eHV system. The post-fault active power in nLVRT mode is even less than slight0,00 slope in steady-state post-fault400, voltage for the nLVRT 0,00 0,00 in OS2. The voltage now drops below 0.8 p.u. in all LV and-100, BM modes is caused by the lack200, of governor models for -100, -100, networks, whereas in OS2 the voltage remains above 0.8 p.u. central-200, power plants and related frequency0,00 drop in the system. -200, -200, 0 1, 2, 3, 4, [-] 5, 0 1, 2, 3, 4, [-] 5, in the rural0 1, LV networks2, 3, connected4, [-] 5, to the0 NE1, system2, 3, nodes4, [-] 5, 20 and 29. Post-fault only 75 MW remains in the nLVRT The active power output by all LV connected DG (PV, mode, meaning 92 percent of pre-fault LV power is lost. The CHP) is shown in figure 11b. Pre-fault power output is equal voltage boosting by the BM in OS3 keeps the voltage above for all control modes at 519 MW. After the initial drop in 0.8 p.u. in the affected distribution systems and thus post- power, the power output rises for all modes except the BM. fault power is higher than in the nLVRT case. After the active For the nLVRT mode, this is caused by the increase of active power is ramped up post-fault in the BM case, the output of current by LV connected PV systems. For the aRCI and aRACI LV connected DG is 639 MW. The aRCI and aRACI modes modes, the injection of additional current in the LV network manage to keep even more old DG online and have a post-fault contributes as well since the PV systems were only loaded power output of 647 MW. half pre-fault. The BM has the largest initial power loss as the The active power ramp of the BM is paused at approxi- active current setpoint is driven to zero almost immediately mately 1.5 s, when the voltage in the LV networks connected after fault occurrence. Post-fault however, the BM almost to eHV busbar 39 drops below the threshold of 0.9 p.u. and returns to the same value as the aRCI and aRACI modes. The the connected LVRT-capable PV systems return to their fault difference arises from the fact that the voltage boosting by the state for a short time. This is also the reason that there are still aRCI and aRACI modes has kept some of the old nLVRT DG some PV systems ramping up their power after 4 s. Connected at the outskirts of the voltage dip online by keeping the voltage to eHV busbar 39 is an equivalent generator, which due to its above 0.80 p.u. during the fault. The BM does not boost the large size causes larger voltage swings here than anywhere voltage significantly in OS2 and thus not all old DGs inject else in the system. power post-fault. The aRCI and BM show the post-fault active power ramp for the PV systems. The aRACI mode does not show a ramp in active power post-fault, as the setpoint during C. Discussion fault is even higher than in steady-state. The total power output Figure 13 shows the total generation by DG in the test post-fault by the LV connected DG is 208 MW for the nLVRT system pre-fault and the lost power post-fault for all oper- mode, 313 MW for BM and 338 MW for the aRCI and aRACI ational scenarios with a k-factor of MV DG of 0.2 p.u. It modes. This translates to a loss of power compared to the pre- is clear that higher in-feed pre-fault leads to a larger loss fault condition of 60, 40 and 35 percent respectively. post-fault. Furthermore, in each OS the power lost is greatest 1,20 1,20 1000, 1000,

DIgSILENT 1,00 1,00 800, 800,

0,80 0,80 600, 600,

0,60 0,60 400, 400,

0,40 0,40 200, 200,

0,20 0,20 0,00 0,00 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5,

1,20 1000, 1000, 80,0

1,00 800, 800, 64,0

0,80 600, 600, 48,0

0,60 400, 400, 32,0

0,40THIS PAPER WAS PRESENTED AT THE 4TH200, SOLAR INTEGRATION WORKSHOP (2014)200, AND PUBLISHED IN THE WORKSHOP’S16,0 PROCEEDINGS. 7 2022 OS3 LV connected DG power 0,202022 OS3 Voltage of eHV-19-2001_HV-03_MV-02_LV-02 0,00 0,00 0,00 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [-] 5, a MV k-factor of 2 p.u. could be an effective measure to 1,201,20 1000,1,20 1000,1000, 1,201000, [p.u.] [MW] further boost the voltage. DIgSILENT 1,001,00 800,1,00 800,800, 1,00 800,

0,800,80 600,0,80 600,A600, minimum requirement of full0,80 dynamic600, voltage support

0,600,60 400,0,60 for400,400, LV and MV connected PV systems0,60 400, would have a high

0,400,40 200,0,40 impact200,200, on the protective system0,40 prevalent200, in the respective LV and MV networks. The protective system might have to 0,200,20 0,000,20 0,000,00 0,20 0,00 00 1,1, 2,2, 3,3, 4,4,[s][s]5, 5, 0 0 1, 1, 2, 2, 3, 3, 4, [s]4, [s]5, 5, be revised0 0 1, in1, 2, order2, 3, to3, 4, prevent[s]4,5,[s] 5, blinding0 and01, false1,2, tripping2,3, 3,4, [s] of4,5, [s] 5, (a) Voltage in busbar eHV-20 HV- (b) Active power output by LV DG. protective devices. Additionally, anti-island detection would 1000,031,20MV-02 LV-02. nLVRT in black, nLVRT80,01000, in black, BM in red, aRCI in have1000,1000, to be revised. Given the limited90,0 80,0 voltage support found BM in red, aRCI in blue, aRACI in blue, aRACI in green. 800,1,00 64,0800, 800,800, 60,0 64,0 green. in this paper a requirement for voltage support through aRCI 600,0,80 48,0600, or600, aRACI600, remains questionable. Future30,0 48,0 research will have to Fig. 12: 2022 OS3 results 400,0,60 32,0400, assess400,400, its value. 0,00 32,0

200,0,40 16,0200, 200,200, -30,0 16,0 3.500 CHP PV Wind 0,000,20 0,000,00 0,000,00 IV. CONCLUSION-60,0 0,00 3.00000 1,1, 2,2, 3,3, 4,4,[s][s]5, 5, 0 0 1, 1, 2, 2, 3, 3, 4, 4,[-] [s]5, 5, 0 0 1, 1, 2, 2, 3, 3, 4, [s]4,5,[s] 5, 0 01, 1,2, 2,3, 3,4, [-] 4,5, [-] 5, Current grid codes in Germany and most other European 2.500 1,20 1000, countries1000, still mandate LV connected DG1,20 to disconnect in case 2.000 1,00 800, of voltage800, dips below 0.8 p.u. The impact1,00 of high penetration 1.500 of PV systems in LV distribution systems following a 0,80Megawatt 600, distant600, transmission system fault has0,80 been investigated. Case 0,601.000 400, studies400, were performed for pre-fault0,60 power flow levels and 0,40 500 200, various200, LV connected PV LVRT control0,40 modes with realistic

0,20 0 0,00 composition0,00 of DG in active distribution0,20 systems. 0 1, CM12, CM2 3,CM3 CM44, [s] 5, CM1 CM2 CM30 CM41, CM12, CM23, CM34,CM4[s] 5, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [s] 5, In- Lost power In- Lost power In- Lost power The simulation results suggest that if current GCR remain 1000, feed feed 80,0 feed 1000, 90,0 unchanged, voltage sags below 0.70-0.75 p.u. retained voltage 800, OS1: OS2:64,0 OS3: 800, 60,0 "warm summer afternoon" "windy cloudy fall day" "windy clear spring day" at transmission level will cause the disconnection of non-LVRT 600, 48,0 compliant600, distributed generation. The30,0 range of the retained Fig. 13: Active power lost in the test system due to the 400, 32,0 voltage400, depends on the voltage support0,00 of other, LVRT-capable transmission system fault with k=0.2 for all ‘new’ MV DG. 200, 16,0 DG200, at distribution level. For this test-30,0 system, a lack of LVRT capability of LV connected PV systems causes a loss of 0,00 0,00 0,00 -60,0 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [-] 5, 0 1, 2, 3, 4, [s] 5, 0 1, 2, 3, 4, [-] 5, in the nLVRT (CM1) case. The BM (CM2) performs better active power of 908 MW in the worst case scenario in 2022. than the nLVRT mode in all scenarios. The aRCI and aRACI Considering the active power ouput of all DG in the test modes (CM3 and CM4 respectively) perform best and lead system, approximately 1 400 MW is lost, equivalent to a large to identical amounts of power lost in this study. Increasing centralised power plant. Realising that the amount of ‘active’ the k-factor of MV DG to 2 p.u. in 2022 OS3 decreased distribution networks in the test system could be more than the disconnected capacity in the nLVRT case the most, by doubled considering the total number of distribution system 24 percent. The effect is greater at higher pre-fault infeed of loads in the eHV level, it can be concluded that the amount of DG and would be substantially larger if also the existing MV active power lost in the system would be significantly higher DG would be set to a k-factor of 2 p.u. if all distribution systems were ‘active’. To maintain the post-fault active power balance as best as The frequency containment reserve of the Continental possible, a minimum requirement for new LV connected PV Europe region, 3 000 MW, has been designed for the loss systems should be to ride through voltage dips caused by of two large power plants of 1 500 MW capacity each. The transmission system faults in BM. This control mode should modelling for the benchmark system used in this study be implemented in the short term. In the long term, the aRCI suggests that loss of LV connected DG could become the mode is advised as it is most successful at boosting the voltage largest contingency in future if current grid code requirements at the connection point of the PV system. will not be changed. However, it should be noted that even the aRCI mode has a limited capability of raising the voltage in the LV network Based on the outcomes of this study, future research could during a fault. The penetration of new PV systems in the be conducted on, e.g., the impact of fault location and type rural LV network is 70 percent, and voltage dips that would on the network response. It would also be interesting to lead to a voltage below 0.75 p.u. would not be raised above investigate different aggregation methods to evaluate the 0.8 p.u. by the aRCI mode. Thus, unless penetration levels impact of the aggregation method on the results. Further areas are increased significantly beyond 70 percent for new PV for study are aimed at a more realistic representation of the systems or old PV systems are retrofitted, no control mode German case, such as using a German transmission system is capable of keeping existing LV connected online. Future model and differentiating for urban areas. research must show whether this is acceptable. Requesting THIS PAPER WAS PRESENTED AT THE 4TH SOLAR INTEGRATION WORKSHOP (2014) AND PUBLISHED IN THE WORKSHOP’S PROCEEDINGS. 8

APPENDIX [7] ENTSO-E. Dispersed Generation Impact on CE Region Security: Dynamic Study: Final Report, 22.03.2013. TABLE V: Typical DG penetration levels in Germany, rounded [8] ENTSO-E. ENTSO-E Network Code for Requirements for Grid Con- nection applicable to all Generators, 8 March 2013. to nearest 5 percent. Percentages expressed as function of [9] IEEE Standards Coordinating Committee 21. IEEE 1547 Standard for the load connected to a given level. ‘All DG’ values include Interconnecting Distributed Resources With Systems: additional DG types (hydro, geothermal) and may therefore be Workshop Meeting Minutes, December 3 - 4, 2013, New Brunswick, NJ. larger than the sum of PV, Wind and CHP. [10] ENTSO-E User Group. Meeting on “Network Code for Requirements for Grid Connection applicable to all Generators” (NC RfG) on 22 Region type 2 Region type 3 November 2012, 17.12.2012. suburban rural [11] FNN. VDE—FNN-Roadmap identifiziert weiteren Entwicklungsbedarf zum Umbau der Netze (VDE—FNN-Roadmap identifies further need in [%] HV MV LV HV MV LV the transformation of the electricity network): Press release, 07.12.2012. All DG 2012 90 105 65 165 205 110 [12] Aidan Tuohy et al. The Impact of Low Voltage Disconnection of Dis- PV 2012 40 55 60 70 95 100 tributed PV on Bulk Electricity System Reliability. In 3rd International Wind 2012 35 30 0 60 50 0 Workshop on Integration of into Power Systems, October CHP 2012 10 15 5 25 25 5 2013. All DG 2022 145 170 110 265 335 175 [13] Kostis Skaloumpakas. Response of Low Voltage Networks with High PV 2022 60 90 100 115 160 165 Photovoltaic Systems Penetration to Transmission Network Faults. MSc Wind 2022 50 40 0 95 75 0 thesis, Technische Universiteit Delft, Delft, 2014. CHP 2022 20 25 5 40 45 10 [14] Holger Berndt et al. TransmissionCode 2007. Netz- und Systemregeln der deutschen ubertragungsnetzbetreiber.¨ Technical report, Verband der Netzbetreiber - VDN – e.V. beim VDEW, August 2007. August JCB. [15] German Government. Verordnung zu Systemdienstleistungen durch TABLE VI: Test system penetration. Percentages expressed as Windenergieanlagen (Systemdienstleistungsverordnung – SDLWindV) function of the load connected to a given level. Bold printed (Ordinance for Ancillary Services of Wind Power Plants (Ancillary Services Ordinance - SDLWindV). Federal Law Gazette, I(39):1734– numbers differ more than 10 percent from the statistical 1746, 2009. analysis results (see table V). Large differences for suburban [16] Federal Statistical Office (Destatis). Regional statistics: Towns and CHP by design. villages. Technical report, 31.03.2011. [17] Deutsche Energie-Agentur GmbH. dena-Verteilnetzstudie: Ausbau- Suburban PV Suburban CHP Rural und Innovationsbedarf der Stromverteilnetze in Deutschland bis 2030. Technical report, Berlin, 11-12-2012. [%] HV MV LV HV MV LV HV MV LV [18] Technische Richtlinien fur¨ Erzeugungseinheiten: Teil 4 - Anforderungen All DG 2012 80 95 60 50 55 5 140 155 100 an Modellierung und Validierung von Simulationsmodellen der elek- All DG 2022 115 135 100 70 65 5 245 285 180 trischen Eigenschaften von Erzeugungseinheiten und -anlagen. Technical PV 2012 35 50 60 5 10 0 65 95 100 report, FGW e.V. Fordergesellschaft¨ Windenergie und andere Erneuer- PV 2022 55 85 100 10 10 0 110 165 170 bare Energien, Kiel, 2011. Wind 2012 35 30 0 35 30 0 55 40 0 [19] Jens C. Boemer et al. Fault ride-through requirements for onshore wind Wind 2022 45 30 0 45 30 0 100 80 0 power plants in Europe: The needs of the power system. pages 1–8, CHP 2012 10 15 0 10 15 5 20 20 0 2011. CHP 2022 15 25 0 15 25 5 35 45 10 [20] Diogenes Molina. Progress report on the development and validation of DIgSILENT PowerFactory model of the 39-Bus benchmark system for stability controls: Addressed to IEEE Task Force on Benchmark Systems for Stability Controls. Technical report, 25-09-2012. ACKNOWLEDGMENT [21] IEEE Society. IEEE recommended practice for excitation system models for power system stability studies. Institute The authors would like to thank FNN for providing the of Electrical and Electronics Engineers, New York and N.Y, 2006. funding of this work. Furthermore, the partners of FNN are [22] Georgios Papaefthymiou et al. Assessment of the Frequency Settings from Distributed Generation in Germany for the Prevention of Frequency thanked for the invaluable information they supplied. Stability Problems in Abnormal System Conditions. 2013. [23] Afshin Samadi et al. Comparison of a Three-Phase Single-Stage PV REFERENCES System in PSCAD and PowerFactory. In 2nd International Workshop on Integration of Solar Power into Power Systems. [1] 50Hertz Transmission GmbH, Amprion GmbH, TenneT TSO GmbH, [24] DIgSILENT GmbH. DFIG Template: Manual. Technical report, and TransnetBW GmbH. Netzentwicklungsplan Strom 2012 (Network DIgSILENT GmbH, Gomaringen, 02.05.2011. Development Plan Electricity 2012): 2. uberarbeiteter¨ Entwurf der [25] DIgSILENT GmbH. Fully Rated WTG Template: Manual. Technical ubertragungsnetzbetreiber¨ (2nd revised draft of the transmission system report, DIgSILENT GmbH, Gomaringen, 02.05.2011. operators). Technical report, 30/05/2012. [26] Edward Jeroen Coster. Distribution grid operation including distributed [2] Deutsche Gesellschaft fur¨ Sonnenenergie (DGS) e.V. EnergyMap.info: generation: Impact on grid protection and the consequences of fault Konsolidierte und plausibilisierte Datenbank der Stammdaten von EEG- ride-through behavior. PhD thesis, Technische Universiteit Eindhoven, Anlagen in Deutschland (Consolidated and plausibilised database of the Eindhoven, 2010. EEG power plant registry in Germany), 2013. [27] Datenblatter¨ zu Generatoren, Erregersystemen, Spannungsreglern, Mo- [3] VDE-FNN. Erzeugungsanlagen am Niederspannungsnetz. Technische toren und Kupplungen: Interne Kommunikation im Rahmen des Projekts Mindestanforderungen fur¨ Anschluss und Parallelbetrieb von Erzeu- - Weiterentwicklung des Verhaltens von Erzeugungsanlagen am Nieder- gungsanlagen am Niederspannungsnetz, 01.08.2011. spannungsnetz im Fehlerfall - Systemsicherheitsaspekte - durchgefuhrt¨ [4] CENELEC. Requirements for the connection of micro-generators in an der TU Delft und beauftragt vom Forum Netztechnik/Netzbetrieb im parallel with public low-voltage distribution networks, 21.12.2010. VDE e.V. (internal project communication). Technical report, Bosch [5] IEEE Standards Coordinating Committee 21. Draft Standard for Inter- KWK System GmbH, 2014. connecting Distributed Resources with Electric Power Systems, June [28] DIgSILENT GmbH. General Load: Technical Reference Documenta- 2013. tion. Technical report, DIgSILENT GmbH, Germany, 2013. [6] Jens C. Boemer et al. Overview of German Grid Issues and Retrofit [29] Jens C. Boemer et al. Response of Wind Power Park Modules in of German Photovoltaic Power Plants for the Prevention of Frequency Distribution Systems to Transmission Network Faults during Reverse Stability Problems in Abnormal System Conditions of the ENTSO- Power Flows: Submitted paper under review. IET Renewable Power E Region Continental Europe. In 1st International Workshop on Generation, 2014. Integration of Solar Power into Power Systems, October 2011.