MODELING ELECTRICITY TRADE in SOUTHERN AFRICA Second Year Final Report to the Southern African Power Pool

March 23, 1999

USAID Co-operative Agreement on Equity and Growth through Economic Research Trade Regimes and Growth (EAGER/Trade)

MODELING ELECTRICITY TRADE IN SOUTHERN AFRICA Second Year Final Report to the Southern African Power Pool

F.T. Sparrow, William A. Masters, Zuwei Yu, Brian H. Bowen PURDUE UNIVERSITY

Peter B. Robinson, Zimconsult

David Madzikanda SAPP-PSC Jean-Louis Pabot SAPP-GPWG Arnot Hepburn Eskom, South Africa B.R.F. Moore BPC, Botswana M. Badirwang BPC, Botswana Fatima Arthur EDM, Mozambique Bongani Mashwama SEB, Swaziland Roland Lwiindi ZESCO, Zambia Edward Tsikirayi, ZESA, Zimbabwe K.R. Abdulla TANESCO, Tanzania

Jose A. Salguerio Energy Sector, TAU, SADC

Swaziland SAPP Management Meeting February 23-25, 1999 2

ACKNOWLEDGEMENTS

Each of the members of Purdue UniversityÕs State Utility Forecasting Group and graduate staff have made significant contributions during the second year of SAPP modeling. These include Doug Gotham, David G. Nderitu, Frank Smardo, James Wang, Kevin Stamber, Basak Aluca, and Pedro Gonzales-Orbago. The administrative assistance from Chandra Allen and Barbara Beaver has been a great help throughout. Thanks is given to each one for contributing so much. Colleagues from the Southern African Power Pool (SAPP) and researchers at Purdue University highly value the funding and support of the USAID/EAGER in making the Purdue/SAPP joint research project all possible.

CONTENTS

Page Acknowledgements...... 2

Section 1 Overview of The SAPP Modeling...... 4

Section 2 Formulation ...... 6

Section 3 Data Collection...... 8

Section 4 Computation...... 15

Section 5 Introduction to the Long-Term Model Runs...... 16

Section 6 Demonstration Model Runs...... 18 6.1 Run One...... 19 6.1.1 Costs...... 19 6.1.2 MW ...... 20 6.2 Run Two...... 22 6.2.1 Costs...... 22 6.2.2 MW ...... 24 6.3 Run Three ...... 25 6.3.1 Costs...... 25 6.4 Run Four...... 27

Section 7 Further Work & Model Implementation...... 29

FIGURES Figure 3.1 ...... 13 3

Figure 3.2 ...... 14 Figure 5.1 ...... 17 Figure 5.2 ...... 17 Figure 6.1 ...... 21 Figure 6.2 ...... 23 Figure 6.3 ...... 26

TABLES Table 3.1 ...... 9 Table 3.2 ...... 11 Table 3.3 ...... 10 Table 3.4 ...... 11 Table 3.5 ...... 12 Table 3.6 ...... 12 Table 6.1 ...... 28 Table 6.2 ...... 28 Table 6.3 ...... 29

BIBLIOGRAPHY...... 30 4

1. Overview of The SAPP Modeling The past two years have seen substantial progress in the modeling of electricity trade in the southern region of Africa. The first year researched the short-term benefits of alternative trading procedures. The second year has seen the construction of a comprehensive (generation and transmission) and transparent long-term planning model which provides insights into the optimal lowest cost construction scenarios for the region over a twenty year horizon. It is hoped that a further year of joint modeling work between the SAPP and Purdue University will see the long- term model effectively implemented and running in at least three or four centers of Southern Africa.

• The first year had studied the short-term benefits of alternative electricity trade contracts within the SAPP. It showed that 80 to 100 million USD could be saved annually from the reduction in production costs as a result of a centralized dispatch method of trading.

• Year two of the modeling of electricity trade in Southern Africa has built upon the earlier work that was completed in year one. The second year is working on the SAPP long-term (LT) model. It is showing that an average regional high growth (5.7%) in electricity demand will cost (for new capacity and operations) the region 23.4 billion USD over the next twenty years. If a regional average low growth (2.0%) in demand in electricity is used then the total cost will be 5.8 billion USD. Expansion in transmission is shown to be very important for the earlier years in the model.

• The second year of modeling has produced a workable and sophisticated LT model that will operate on a personal computer. The data will need sensitive detailed double checking by the SAPP colleagues in the third year of implementation. Changes in data for demand growth, project costing, levels of national independence of electricity supply, capital recovery factors (CRF) and several other variables will significantly alter the outcome from this fascinating and powerful LT model. A halving of the CRF for example in the major hydropower countries of SAPP creates savings of over 1.5 billion USD.

The long-term modeling is largely based on the assumption of a medium or SADC average weighted demand growth rate in electricity (3.8%) over the next 20 years. Regional costs for 20 years amount to $11.4 billion USD with a free trade scenario. With a great concerted effort the costs and technical data in this LT model have been updated in a relatively short period of about six months. This has been achieved through excellent collaboration between the SAPP Generation Planning Working Group (GPWP) and the staff at Purdue University. No commercial software is yet available to achieve the level of integration in utility planning and electricity trading for the long-term which this model provides. 5

The medium demand growth figures (Table 3.5) and many other inputs to the model will require further reassessment over the next several months before the LT model can be used for decision analysis in prioritizing construction projects. Numerous runs of the LT model have been made since the Cape Town modeling workshop (July 1998). Improved constraints and equations have since been added. A superior, working model is now available for SAPP to thoroughly test and use within the region. In some countries a low electricity demand growth rate could be considered as being more appropriate. This decision together with other economic and financial data inputs must be decided upon. A closer scrutiny of all the data has yet to be completed in the region. This second year of modeling has established the working LT model and has shown that it is possible to run such a complex model on a personal computer. The PC platform had been an objective specified by the SAPP management in March 1998.

This year two final report needs to be read in conjunction with two other SAPP modeling documents that are dated January/February 1999. These are: • The SAPP Long-Term Modeling USER MANUAL (Second Edition) [6], • Modeling Electricity Trade in Southern Africa 1999/2000 Ð Year 3 Proposal [13].

The User Manual gives detailed documentation on the modeling work that has been achieved over the past 12 months as well as practical instructions on how to run the model. This manual was requested by the SAPP colleagues who were present at the Cape Town workshop.

The year three proposal gives outlines, ideas and tentative plans on how a year of sensitivity analysis and implementation of the LT model at Purdue and within Southern Africa can be assisted and promoted. The second year has seen a working model produced but extensive sensitivity analysis will be needed in a third year prior to full-scale adoption in regional planning. Institutional outreach will be a major thrust for a third year of work. Institutions (utilities, SADC TAU, regulators, and university institutions) should take an active part in testing the sensitivity of the model. The Purdue modeling team will assist the region in the setting up of the LT model (hopefully at a number of venues), in training, for the interpretation of results, and in further model refinements.

This report outlines the modeling and computational issues that have been resolved and gives an overview of the initial trends in the behavior of the LT model. A brief outline to a third year is included. A highly detailed sensitivity analysis of the meaning of the results from the LT model has started to gain momentum and with further time available, this will be completed. The SAPP utilities also need time for testing the LT model for themselves with technical support from Purdue. The coded files of the model were sent to SAPP management in early February 1999. Two utilities in the region have their PC ready and are 6

getting equipped with the solvers (GAMS and CPLEX). Regular updates of the model can be sent to SAPP or at agreed times.

The long-term model, which SAPP requested to be built, for the second year of modeling, has become very complex. Its increased complexity, compared to the short-term model, arose from there being many more constraints and its having greater computing needs. It has taken all of the second year to make it become an acceptable working model. The workshop, which was conducted in Cape Town, demonstrated how the initial LT model worked in its very early stages. Its effectiveness and structure was discussed at good length. Recommendations were also made at this workshop and have since been implemented. After the workshop the SAPP management instructed the Purdue team that all future discussions and data collection for the long-term model should be conducted primarily through the SAPP generating and planning generation group (GPWG). This proved to be very helpful and time saving from the point of view of both data collection and editing of the first edition of the User Manual.

The second year of modeling is summarized under the following section headings: Section 2 Formulation and User Manual, Section 3 Data Collection, Section 4 Computation, Section 5 Introduction to the Long-Term Model Runs, Section 6 Demonstration Model Runs Section 7 Implementation and Further Work.

The user manual describes in detail all of the modeling work that has been achieved in 1998/99. Further copies of this can be obtained from [email protected] . Please email Ms. Chandra Allen and she will electronically send out a copy of the manual or if preferred it can be sent via regular air mail.

2. Formulation The long-term model is a mixed integer mathematical program which, using the GAMS and CPLEX solver software, minimizes the total costs (capital, fuel, operational and maintenance, and unserved energy) for the capacity expansion of the SAPP up to the year 2020.

The long-term model includes decision variables Ð Òto build or not to build?Ó Òto expand or not to expand?Ó Òto make or buyÓ

Involving: • 600 integer variables 7

• 500,000 continuous variables • 20,000 constraints

The objective function of the long-term model is to minimize the costs of generating and transmission capacity expansion in the SADC region over a 20-year time horizon:

(a) If SAPP were to choose the generation/expansion plan to minimize total SAPP costs.

(b) If each country chooses to meet domestic demand only from domestic power sources.

LT Model Objective Function

Minimize Total Cost (present value)

{ ∑ Capital Costs of Extensions & New sites (Large Coal, Small Coal, Combined Cycle, Hydro, Pumped Hydro, Transmission lines)

+ ∑ Fuel Costs + ∑ Operations and Maintenance Costs + ∑ Unserved Energy}

Many user options are allowed in the model: A few of the major options available include: • Planning horizon options • Reliability options • Demand characterization options • Financial options (cost of capital etc) • Supply options

Full details of the formulation are found in the Long-Term Model User Manual (Second Edition, January 1999). It is an appendix to this Year 2 Final Report.

Most of the issues that SAPP delegates specified at the Cape Town meeting have now been incorporated into the model. These included the following: • Require the discount rate (ÒintÓ) in the objective function to be the same as the cost of capital in the CRF function. • Allow CRF to be function of the country and the lifetime of the equipment, e.g., CRF(z,equip type) • Allow demand growth rates to vary by period, as well as country • Allow unforced outages for thermal on all days except winter peak day • Allow unforced outage for hydro in dry season only 8

• Allow pumped storage and pumped storage expansion in the model • Make UFOR for new plants to reflect yearly % • Allow independent 24-hour patterns for 6 day types rather than scalar mult • Allow dgrowth to be indexed by day type, as well as country • Allow yearly peak demand to be obtained from 24-hour table, not PeakD(ty,z) • Allow DSM(z) to enter by altering 24-hour load shape as well as subtracting from demand; allow DSM(z,th) costs to enter • Make extension of Kariba S. conditional upon start of Batoka, not vice versa (add Y(ks) ≤ M Y( B) • Derate all capacities by FOR • Delete INGA III (3,500 MW) and insert GRAND INGA (39,000 MW) • Zim Kariba capacity will be increased to 750 MW by 1999 • Equal share of Kariba water energy allocated to Zambia and Zimbabwe • Reduce the integers associated with new hydro; e.g., Grand Inga only for years after 2010, etc. • All other hydro projects completed by 2000 should be treated as old -- e.g., not in optimization • Require all additions to new and old plants to be multiples of plans of a fixed size -- e.g., new thermal capacity can be 0MW, 300MW, 600MW, 900MW, etc. -- rather than continuous • Allow fixed costs to vary by country, as well as plant type • Allow first capacity increment and cost to differ from additional capacity and cost increments for new plants • Allow NSA to add more than 2 plants/types per period; allow more large coal plants - - up to 8 • Allow plants under construction to enter in year of completion as old plants (no fixed cost); no thermal plants in this situation • Allow for decommissioning/derating of old plants • Derate capacity limits by (1-FORPGO) and (1-FORNα) • Add NSA mothballed plants as new plants with fixed start-up cost • Allow heat rate for large coal and small coal plants to be different • Require additions to new and existing line capacity to be multiples of a fixed capacity • Allow Angola-Zambia, Nam-Zim • Make clear line resistance considered in calculating line loss • By year 2000, assume interconnectors between Zam and Tan, Zam and Mal, Moz and Mal

Initial runs of the LT model in January/February 1999 consider two scenarios in the SAPP. One is with each country having autonomy factors (AF) of effectively 100% and the second where imports are more liberalized. The constraint in the formulation that reflects this change is expressed as: Total generation in country z > Total Demand in country z (AF) 9

3. Data Collection The initial or existing data for generation and transmission, in the long-term (LT) model, includes all of the data from the short-term model plus that for Angola, Malawi and Tanzania. Some generating and transmission expansion plans are currently being constructed and so these are also included in the LT model as existing. The starting year for the LT model is 2000. All generation and transmission lines that are to be in operation in the year 2000 are considered as existing and these are shown in Tables 3.1, 3.2 and in Figure 3.1. Over the past few months all data insertions and corrections have been based on the official (GPWG) SAPP spreadsheets. In some cases these sheets need to be updated and completed but Purdue has honored the request that all data should now be based on data from these spread-sheets.

Table 3.1 SAPP Thermal Generation Data for Existing Plants at Year 2000

Country & Pgmax Country & Station Name (MW) Station Name PGmax (MW) Angola 113 Botswana 118 RSA Lesotho 2 Arnot 1320 Mozambique Duvha 3450 Beira 12 Hendrina 1900 Maputo 62 Kendal 3840 Kriel 2700 Namibia Lethabo 3558 Vaneck 108 Majuba 1836 Paratus 24 Matimba 3690 Matla 3450 Zimbabwe Tutuka 3510 Hwange 956 Koeberg * 1840 Munyati 80 (* Nuclear) Harare 70 Bulawayo 120 Tanzania GT10 37 Swaziland 9 LM6000 75

The generation capacity expansion options, proposed by the individual SAPP utilities, are listed in Table 3.3. Besides these options, there are also some generic generating options that the LT model will consider. These are listed in Table 3.4

Figure 3.1 shows the international transmission network that is expected for the year 2000 and Table 3.6 shows the new international lines that could be considered as options for the future. Once a line is in existence then the LT model is allowed to expand that 10 existing line. It is necessary to ensure that each of the three new countries on the grid (Angola, Malawi and Tanzania) can participate in the trade of the LT model.

Beyond the year 2000 the electricity demand growth figures will determine the magnitude of the increases in capacity. The initial runs of the LT model are using the medium scenario values that are listed in Table 3.5.

The accuracy and reliability of the data can have very significant effects on the output from the model. Whether a new hydropower includes the cost of the dam in the first stage of construction or whether this is averaged over the whole project will be a significant factor that will allow the model to prefer one option over another. 11

Table 3.3 New Generating Options (A Revision of Year 2 Interim Report Appendix V, February 12, 1999) Country Powerstation Num Unit Total Cost Type Cost Heat Fuel O & M H = Hydro ber of Size Added $ T/H/ $/kW Rate Cost Cost T-SC = Small coal Units (MW) MW) millio PS Btu/ $/ $/MWh T-LC = Large coal n KWh MWh Optional Projects Angola Cambambe (Ext.) 2 45 90 H Angola Capanda II 2 130 260 300 H

Botswana Moropule 2 120 240 420 T-SC

DRC Inga3 3500 4900 H DRC Grand Inga ST1 8 750 4750 9780 H DRC Grand Inga ST2 9 750 4750 6330 H DRC Grand Inga ST3 9 750 4750 6120 H DRC Grand Inga ST4 7 750 4750 6070 H

Malawi Kaphichira Phase1 * 2 32 64  H Malawi Kaphichira Phase 2 2 32 64 24 H Malawi Lower Fufu 2 45 90 119 H Malawi Mpatamanga 5(?) 63(?) 315 335 H Malawi Kholombidzo 2 35 70 330 H

Mozambique Mepanda Uncua 5 400 2000 2800 H Mozambique Cah. Bassa N. (Ext.) 1240 1054 H

Namibia Kudu (Gas) 1 750 750 650 T Namibia Epupa 1(?) 360 360 408 H

South Africa Komati A (Recomm). 5 5x90 450 T-SC South Africa Grootvlei (Recomm). 6 5x190+1 1130 T-SC x180 South Africa Komati B (Recomm). 4 4x110 440 T-SC South Africa Camden (Recomm). 8 190 1520 T-SC South Africa PB Reactor 10 100 1000 T-SC South Africa Lekwe 6 659 3950 T-LC South Africa Pumped Storage A 3 333 999 PS South Africa Pumped Storage B 3 333 999 PS South Africa Pumped Storage C 3 333 999 PS South Africa High head UGPS 2 500 1000 PS South Africa Gas Turbine 4 250 1000 T

Tanzania Ubungo (Gas) 1 40 40 24 T Tanzania Tegeta (Gas) 10 10 100 90 T Tanzania Kihansi 3 60 180 270 H Tanzania Rumakali 3 74 222 439 H Tanzania Ruhudji 4 89.5 358 515 H

Zambia Itezhi-Tezhi 2 40 80 74 H Zambia Kafue Lower 4 150 600 450 H Zambia Batoka North 4 200 800 898 H Zambia Kariba North (Ext.) 2 150 300 223 H

Zimbabwe Hwange Upgrade 1 84 84 130 T-SC Zimbabwe Hwange 7 & 8 2 300 600 554 T-SC Zimbabwe Gokwe North 4 300 1200 960 T-SC Zimbabwe Batoka South 4 200 800 898 H Zimbabwe Kariba South (Ext.) 2 150 300 200 H

Commited Projects South Africa Majuba* 6 3x612, 3837  T 3x667 South Africa Arnot 3-6 Recomm.* 4 4x330 1320  T * Commissioned by 2000, **Estimated, *** Estimated by given 664 and assumed without dam. 12

Table 3.2 SAPP Hydropower Generation Data for Existing Plants at Year 2000

Country & Hmax Country & Hmax Station Name (MW) Station Name (MW)

Angola 121 Malawi 56 Nkala 124 37 Tedzani I&II 40 37 Tedzani III 60 37 DRC RSA Inga 1775 Gariep 320 Nseki 248 Vanderkloof 220 Nzilo 108 Palmiet # 400 Mwadingusha 68 Drakensbur# 1000 Koni 42 Tanzania Lesotho Kidatu 200 Muela 72 Mtera 80 Pangani 68 Mozambique Mumgu-Hale 29 HCB 2075 Chic-Cor-Mav 81 Zimbabwe SMoz 16 KaribaSouth 666 Zambia Namibia KaribaNorth 600 Ruacana 240 Kafue 900 Victoria 100 Swaziland 40 (# Pumped)

Table 3.4 SAPP Countries/Nodes that are Allowed New Thermal Generic Sites

1 2 3 4 5 6 7 8 9 10 11 12 13 14 An Bots DRC Les Mal Nam Nmo SMoz NSA SSA Swaz Tan Zam Zim g z NSC • • • • • • • • • • NLC • CC • • • • • GT • • • • • • • • • • • • • • NSC = New Small Coal CC = Combined Cycle NLC = New Large Coal GT = Gas Turbine 13

Table 3.5. Electricity Demand Growth Rates for 1996 to 2020

COUNTRY LOW MEDIUM HIGH % p.a. % p.a. % p.a.

Angola 6.2% 7.9% 10.5% Botswana 3.7% 4.3% 6.0% Lesotho 3.4% 6.1% 8.2% Malawi 2.1% 3.2% 6.2% Mozambique 8.9% 10.9% 13.1% Namibia 5.7% 7.2% 8.5% South Africa 1.8% 3.6% 5.5% Swaziland 2.5% 3.4% 4.5% Tanzania 4.0% 6.0% 7.7% Zambia 2.7% 4.4% 6.4% Zimbabwe 2.9% 4.1% 5.9% SADC tot/weighted av 2.0% 3.8% 5.7% South Africa 1.8% 3.6% 5.5% Rest of SADC 3.6% 5.0% 6.9% Source: Cape Town Modeling Workshop - P.B.Robinson, July 1998

Table 3.6 SAPP International Transfer Capacity Options for 2000 - 2020

Ang Bots DRC Les Mal Nam NMoz SMoz NSA SSA Swaz Tan Zam Zim Ang 2000 2000 Bots DRC 2000 1000 Les Mal 240 Nam 2000 Nmoz 1600 Smoz 1600 NSA SSA Swaz Tan Zam 1000 240 Zim Source: SAPP_Purdue Modeling Workshop, Cape Town, July 1998 14

FIGURE 3.1 SAPP International Maximum Practical Transfer Capacities Existing or Committed for the Year 2000 (MW) [Supplied by Eskom, December 1998]

10 320 9 180 1 4 11 1200 300 550 5B 12 650 2000 2 5A 6 1400 850 700 7A 1400 1200 3500 130 8 7B 3

1. Angola (H) 4. Malawi 7A. N. South Africa (T) 10. DRC 2. Botswana (T) 5A. S. Mozambique (H) 7B. S. South Africa 11. Zambia (H) 3. Lesotho 5B. N. Mozambique 8. Swaziland 12. Zimbabwe (H, T) (H) = hydro site 6. Namibia (H, T) 9. Tanzania (H, T) (T) = thermal site 15

Figure 3.2 The Files that Comprise the Long-Term Model

Main.lst Generic output created by Main.gms Main.gms

Data.inc Lines.inc

Param.inc Reserve.inc

Hydro.inc Uncertain.inc

Thermo.inc Sixhr.inc

Output.inc

Therm_exp.out Hyd_exp.out Cost.out

Trade.out Trans_exp.out 16

4. Computation It is very important to remember that the greater the accuracy of the model then the length of time taken to run the model will also be much longer. Fast estimates can be obtained from the model by setting the accuracy at 3% or 4% (OPTION OPTCR = 0.03 or 0.04).

The tolerance or accuracy of the model is normally set at a very fine value. A default value of 0.000000002%. is used. This can be easily changed by entering the Main.gms file and changing the OPTION OPTCR value at the top of the file. A Ò1% accuracyÓ would be given the value 0.01 and 0.1% given 0.001 etc.

It is a major decision in the LT model on whether to allow a variable to be discrete (binary or integer) or of continuous type. Changes in these values will significantly change the total running time of the model. Any or all variables can be chosen to be either continuous or discrete. The following variables are the most likely, however, to be changed if any:

Yh(ty,z,nh), Decision (binary) variable for a new hydro plant Ypf(ty,z,zp), Decision (binary) variable for a new transmission line YCC(tya,z,ni), Decision (binary) variable for a new combined cycle plant YLC(tya,z,ni),Decision (binary) variable for a new large coal plant

If the binary variable is zero then no construction takes place. When it is one, then construction will take place.

The effect of switching one or more of these variables between continuous and discrete has been investigated through several experiments and the model was initially run with all variables set to continuous. Great caution is advised in making these changes as dramatic increases in model running time may result. It is recommended to consult with SUFG staff (email [email protected]) before changing the base model types.

When the variables Yh(ty,z,nh) in file main.gms, section 3, are set to be discrete, then the run time considerably increases. Switching Ypf(ty,z,zp) in the main.gms file, section 4, does increase the run time as expected but the increase in the run time is not as important a concern as with the possible advantages from this operation. As expected, switching Yh(ty,z,nh) and Ypf(ty,z,zp) together to discrete variables does increase the run time, but this increase is very small compared with increase observed when only Yh(ty,z,nh) is changed. Switching only YLC(tya,z,ni) in the main.gms file, section 6, increases the run time, and this increase is also expected and reasonable 17

Running the model requires the use of GAMS and CPLEX software. At the March 1998 request of the SAPP management it has been made possible for the LT model to run on a personal computer. The specification for the PC is below:

PentiumII BX 100 MHz motherboard, PentiumII 350 MHz processor, 512 Mb 100 MHz RAM, 961Gb UW SCSI hard drive.

Early runs on an IBM laptop computer are proving successful when using a processor speed of 300 MHz. The specification for this is: IBM ThinkPad 770 PentiumII 300 MHz 8.1 GB hard drives, 320MB RAM, Win NT

All of the model files are shown in Figure 3.2. All of the equations and constraints are in the Main.gms file. There are nine input files and six output files. The LT model user manual gives detailed explanation on how to operate the model.

5. Introduction to the Long-Term Model Runs The second year of SAPP modeling produced a comprehensive and transparent working LT model which minimizes the total expansion costs (fixed and variable) for the whole Southern Africa region. A twenty-year time horizon is used most of the time.

January/February 1999 is seeing the model function most effectively and it will now be essential for the regional experts to confirm the data in the files. Runs at the end of 1998 indicated that the major new hydropower station at Inga (Grand Inga ST1) would be chosen by the model. With the modification to the Inga cost data however in January 1999 (cost increase) the project is not chosen at all now by the model. It is therefore vitally important that a thorough inspection of all data in the model now takes place similar to the testing rigors that the formulation has been and continues to go through. Major decisions can be influenced by the model and it is necessary that the region be given time to check all of the model inputs as part of the proposed third year of model implementation.

When the SAPP countries relax their autonomy constraint (and there is an average electricity demand growth) then the regional costs amount to 11.47 billion USD over the twenty years 2000-2020. Figure 5.1 shows how the bulk of these costs arise from the 18 construction of expansion of thermal power stations. The majority of the thermal power is built in the second half of the 20-year horizon as shown in Figure 5.2.

When the autonomy constraint is enforced then the total regional cost increases to 12.17 billion USD, an increase of just over 6%. The extra costs occur from increased thermal construction in the earlier years. An analysis of these results now follows. 19

Figure 5.1 Total Expansion Costs for Each Period for Thermal and Hydropower Generation and Transmission. Expansions - with Relaxed Autonomy Constraint in the 20 Year Horizon (Millions of USD per Period) (Total 20 year cost is 11.4 billion USD with average growth)

2000 1800 Thermal 1600 1400 Hydro 1200 1000 Trans 800 Total 600 400 200 Expansion Costs (Million USD) 0 2005 2010 2015 2020 Time Horizon (Years)

Figure 5.2 Thermal and Hydropower Expansions for Each Period with Relaxed Autonomy Constraint in the 20 Year Horizon (MW per Period.) (Total 20 year cost is 11.4 billion USD with average growth)

12000

10000

8000 Thermal 6000 Hydro

4000 (MW per Period) 2000

Total Capacity Expansions 0 2005 2010 2015 2020 Time Horizon (Years) 20

6. Demonstration Model Runs

In order to demonstrate the usefulness of the model in a range of applications, model runs were performed to suggest answers to several fundamental questions of importance to SAPP members:

• What mix of projects currently proposed by SAPP members would be chosen, if SAPP’s goal was to minimize the near term (10 year horizon) present value of construction and operating costs without regard for country autonomy, considering only the required reserve margins of the system?

• In the longer term (20 year planning horizon), what is the economic cost to SAPP members of self sufficiency, defined to require that each country maintain sufficient domestic generating capacity to meet their peak demands without relying on firm imports?

• How would the SAPP wide least cost expansion plan be altered if SAPP utilities had access at competitive prices to “off the shelf” generation technologies (gas turbines, combined cycle, small coal, large supercritical coal units) sold worldwide?

Figures 6.1, 6.2, and 6.3 summarize the results of these demonstration runs in general terms, without identifying specific projects. The reason for this lack of specificity is not because the model doesnÕt produce specifics, but rather to avoid the specific results being used improperly or prematurely to either promote, or discourage the funding of specific projects by outside agencies. Such use of the model must wait careful checking of the database and model structure to insure both meet SAPP intentions and specifications. One of the great benefits of the modeling structure is its ability to create a Òlevel playing fieldÓ, on which the various SAPP generation and transmission projects can compete for inclusion in the least cost plan. However, at the present time, the model structure is not yet fully tested, nor more importantly, the economic/engineering database checked for consistency, to allow such comparison. More specifically, while the generation expansion options available for selection (particularly when augmented with generic options based on representative U.S. data) are adequate, the same cannot be said for the transmission expansion options. It appears that SAPP has so far devoted less time to developing these options, than to the set of generation options. Consequently, the results are presented in summary form only, without identifying the specific projects in the various solutions; that must await a thorough checking of the model and its supporting database.

Four runs are considered: Run One Ð Eight year horizon and a relaxed autonomy constraint Run Two Ð Twenty year horizon and a relaxed autonomy constraint 21

Run Three Ð Twenty year horizon with autonomy enforced Run Four Ð The impact of demand forecast error

6.1 Run One: The Midrange Cost Minimizing Mix of proposed SAPP Projects, with the country autonomy constraint relaxed. The scenario description follows.

Scenario Description This scenario seeks the least cost mix of SAPP projects under the following assumptions: • The generation and transmission capacity expansion options are limited to those proposed by SAPP as described in Table 3.3 and Figure 3.1. • Capital recovery factors are set at 12% for hydro and transmission line projects, 15% for thermal plants. • The planning horizon is 8 years; consisting of choices in 2002, 2004, 2006, and 2008, with SAPP demand growing at the average (3.8%) yearly growth specified by Dr. Peter Robinson. (3.8%) is the average of 11 SAPP country growth rates, which vary from 1.8%/year to 8.9%/year). • Initial (2000) capacity includes existing capacity, plus those committed projects approved at the Cape Town meeting in July 1998, and modified since then. • The country autonomy constraint, which would require that a country construct enough domestic generation capacity to meet peak requirements without relying on imports, is relaxed. • Each country is only required to maintain a reserve margin (met by either domestic capacity or firm imports) constraint with SAPP guidelines (19% for thermal, 10% for hydro). • Initial Fuel and O & M operating expenses are where possible, taken from SAPP documents (coal $.30 to $.60/106 BTU: gas, $1.10 to $4.00/106 BTU). • Such costs escalate at model default values (1%). • Water cost $1.50/MWh. • Unforced outage rate, 7 to 11%, depending on the equipment. • Forced outage rates, 3 to 10%, depending on the equipment. • O & M in the $.0006 to $.005/Kwh range. • Transmission losses proportional to distance.

The results of the scenario run are shown in Figure 6.1. The numbers are the MW of Thermal and Hydro (circled at the nodes) and Transmission (boxed along the arcs) capacity installed over the 8 year horizon. Several characteristics of the solution stand out. 22

6.1.1 Costs • Almost 4/5 of the total cost is in the variable cost of operation of existing and new generation units, as the system chooses to utilize regional existing capacity to the fullest extent possible, before constructing new facilities.

• Variable costs are dominated by fuel cost (75%) followed by O & M (16%) and water use charges (10%) (This is somewhat arbitrary, since no consensus exists on proper opportunity costs for water; the model uses $1.50/MWh).

• Thermal and Hydro expansion costs (2/3 of total expansion cost) are fairly evenly spread over the 8 year horizon, while transmission cost (1/3 of expansion costs) are concentrated in the beginning and end of the planning period.

6.1.2 MW • Substantial (40%) resources are devoted to improving the transmission system. Early in the planning period, this involves increasing the capacity of the DRC/Zambia/Zimbabwe/RSA ÒspineÓ of SAPP to allow the substitution of cheap excess capacity in the North to forestall the need for new construction in the South. Towards the end of the planning period, the model invests in an alternate North-South route through Angola, Zambia, and RSA to create a competitive path for electricity to flow from North to South.

• New hydro construction represents almost three-quarters of new MW construction - 2860 of the 4700 MW the model calls for. Of this total, over half (2000 MW) is devoted to constructing new pumped hydro facilities, which in effect, allows thermal plants to mimic the energy storage advantages of standard hydro facilities.

• The remainder of the hydro investments are along the two ÒspinesÓ of the SAPP system, involving hydro construction in Zambia and Zimbabwe early in the planning period, and Angola and Namibia towards the end.

• Thermal construction is limited to the recommissioning of a substantial amount of currently mothballed small coal capacity, whose cost/KW is extraordinarily inexpensive when compared to other thermal capacity, and construction of gas turbines. As a result, while thermal capacity additions represent about 40% of total new MW generation installed, they represent less than 1/10 of total expansion cost.

To repeat; these conclusions are completely dependent upon the data inputs into the model as given in Table 3.3. Proposal projects now excluded in the solution will enter the 23 solution if their capital and operating expenses fall below that of the projects selected by the model.

Finally, it should be emphasized that these results presume that each countryÕs natural need for autonomy, in the form of construction for sufficient reserves to allow peak demands to be met by domestic capacity alone, has been suppressed. The result would look quite different if such constraints were added. 24

Figure 6.1: Expansion Results of 8-year Horizon Model, Country Autonomy Constraint Relaxed (all units on map in MW)

1,223 NH-149 2,000 NH-130 108

NH-32 2,000 500 NH-190 300

1,085

586

NH-360

SC-1590 GT-250 NH = new hydro 321 SC = small coal PH = pumped hydro CC = combined cycle 1,172 GT = gas turbine LC = large coal = new transmission capacity PH-1999 = new generation capacity

Values are in Billions of Dollars Total Variable Costs for the Horizon Total Expansion Costs for the Horizon Fuel 3.217 Thermal 0.112 O&M 0.703 Hydro 0.555 Water 0.411 Transmission 0.437 Unserved En. 0.000 Total 4.331 Total 1.104 Total: $5.435 Billion *** DEMONSTRATION RUN ONLY Ð Not to be used for project evaluation *** 25

6.2 Run Two: Long Term Capacity Expansion Options for SAPP with the Country Autonomy constraint relaxed.

Scenario Description The following assumptions were made in this demonstration run: • The planning horizon was over 20 years, allowing choice to be made in 2005, 2010, 2015, and 2020; SAPP demand growth was again the average scenario specified by Dr. Robinson.

• The set of capacity expansion options was expanded to include the construction of generic plants, whose cost and performance data were taken from a recent U.S. study.

• The country autonomy constraint remained relaxed, only requiring that SAPP countries satisfy their reserve requirements by a mixture of thermal and hydro capacity and firm imports.

• All other assumptions were the same as those used in Run One.

The results of this scenario are shown in Figure 6.2. Again, new generation capacity is circled, new transmission capacity is boxed.

6.2.1 Costs • In contrast to the short-term case just described, expansion costs are a substantial (40%) portion of total system costs. This is to be expected, as system demand catches up, and then surpasses system supply, over the longer planning horizon.

• Fuel costs again dominate operating costs.

• Thermal and hydro costs are a much larger fraction (85%) of total expansion costs; transmission costs represent only 15% of total expansion costs. 26

Figure 6.2: ExpansionResults of 20-Year Planning Horizon, Country Autonomy Constraint Relaxed (all units on map in MW) NH-3500 3,726 CC-1046 2,000 NH-262 CC-152 GT-143 138

NH-120 2,000 500 NH-557 300

2,750 NH-572 GT-1320 SC-1800 GT-840 2826

NH-360 GT-235 CC-485 CC-3552

LC-5615 NH = new hydro SC-1590 43 SC = small coal GT-1222 CC-406 PH = pumped hydro GC = combined cycle 972 GT = gas turbine GT-343 LC = large coal LC-1372 PH-3000 = new transmission capacity CC-6277 = new generation capacity

Values are in Billions of Dollars Total Variable Costs for the Horizon Total Expansion Costs for the Horizon Fuel 5.604 Thermal 2.651 O&M 0.923 Hydro 1.187 Water 0.419 Transmission 0.690 Unserved En. 0.000 Total 6.947 Total 4.527 Total: $11.474 Billion *** DEMONSTRATION RUN ONLY Ð Not to be used for project evaluation *** 27

6.2.2 MW • The vast majority of additional MW is devoted to new generation construction, rather than transmission. This is not because transmission expansion substantially decreased as the horizon lengthened, but rather because of the substantial increase (5 fold!) in the need for new generation, as the planning horizon extended into the years where substantial generation deficits exist.

• The general pattern of transmission capacity expansion found in the short- term model persists in the longer-term model Ð initially invest in the DRC/Zambia/Zimbabwe/RSA ÒspineÓ, then build the western route from DRC through Angola and Namibia to RSA.

• In contrast to the short-term model, thermal construction becomes a major (50%) part of the total construction dollar budget, and an even larger (over 70%) part of the MW total. Thermal construction is dominated by construction of over 6000 MW of large coal generation plants in the South, and a surprising amount of construction of generic combined cycle plants in those countries where a source of natural gas supply was identified. In addition, relatively small investments were made in small coal plants, and generic gas turbines for peaking purposes, again restricted to those countries where gas was assumed available at competitive prices.

• New Hydro construction was dominated by drawing on the resources of DRCÕs Inga site towards the end of the horizon; early hydro expansion was, as in the short-term model, concentrated in hydro investments along the DRC/Zambia/Zimbabwe/RSA ÒspineÓ, and in pumped storage investment in the South.

These results should be viewed with even more caution, if possible, then the results of the previous run. They assume SAPP has access to thermal technologies at the going world average cost. This may be a wildly optimistic assumption, given the peculiar nature of generation plant construction in the SAPP region. In particular, it assumes that the SAPP region can invest in the substantial infrastructure necessary to allow the bulk (all countries except Zambia, Malawi, Mozambique, Swaziland, and DRC) of SAPP members to have access to natural gas as a fuel source, either from domestic production, or imports. Such access would allow SAPP members to benefit from the substantial advantages of gas fired combined cycle generation plants, which now dominate new construction in North America. (The assumption is easy to change by simply eliminating generic combined cycle plants from consideration.) 28

6.3 Run Three: Long-Term Capacity Expansion Options for SAPP, with the country autonomy constraint enforced.

Scenario Description The only change from Run Two is to enforce the constraint that, in addition to satisfying SAPP wide reliability requirements, each country chooses to construct enough domestic generation capacity to allow its domestic peak demand to be satisfied by domestic capacity alone. This does not mean that trade will not take place; it simply raises the economic hurdle for imports to require that the full cost of imports compete with the out of pocket (marginal) cost of producing power from existing domestic “excess” capacity.

The results of this scenario are shown in Figure 6.3.

6.3.1 Costs It is somewhat surprising to note that the requirement that counties maintain sufficient domestic capacity to meet peak demands without dependence on imports increases the total SAPP cost by only 8%, about the magnitude of the increase seen in the short run model.

This, like all the other results, must be viewed as provisional, until enough experience is gained through use of the model to feel confident in the results. Nonetheless, it is interesting to conjecture why the unexpectedly small difference could take place;

• Trade volumes with the constraint relaxed were not that big in the first place.

• The Òmake vs. buyÓ decision was not that much affected by the presence of Òback-upÓ domestic capacity, since the marginal cost of imports was below the out-of-pocket cost of domestic production.

• The 20-year model allows all countries access to the same U.S. generic power plant designs, thus limiting regional differences in generation cost, which provide the impetus for trade in the first place.

As in the previous cases, extreme caution should be used in interpreting these results. It is extremely doubtful that all countries will insist on 100% domestic generation coverage of their domestic peak demands. Thus, this run should only be used to indicate the rough order of magnitude of the cost increases when nations insist on self-sufficiency. 29

Figure 6.3: Expansion Results of 20-Year Planning Horizon, Country Autonomy Constraint InForce (all units on map in MW) NH-3500 3,726 CC-1269 2,000 NH-188 CC-152 GT-143 214

NH-166 2,000 500 NH-558 300

2,675 NH-564 GT-193 SC-1800 GT-976 2648

NH-360 GT-157 CC-353 CC-3552

LC-5937 NH = new hydro SC-1590 43 SC = small coal GT-1900 CC-406 PH = pumped hydro GC = combined cycle 1009 GT = gas turbine GT-173 LC = large coal LC-1131 PH-3000 = new transmission capacity CC-5756 = new generation capacity

Values are in Billions of Dollars Total Variable Costs for the Horizon Total Expansion Costs for the Horizon Fuel 5.562 Thermal 3.359 O&M 0.915 Hydro 1.125 Water 0.425 Transmission 0.690 Unserved En. 0.000 Total 6.902 Total 5.270 Total: $12.172 Billion *** DEMONSTRATION RUN ONLY Ð Not to be used for project evaluation *** 30

6.4 Run Four: The impact of demand forecast error on SAPP costs.

Scenario Description: Predicting regional electricity growth is a risky proposition anywhere where it is practiced. The costs of underestimating demand growth require that the system pay the extra costs involved with the construction of units with short construction lead times, as the system tries to rapidly make up the generation supply deficiency, as unexpected higher demands develop. Conversely, the costs of over-estimating demand growth involve the costs of canceling unnecessary projects planned in anticipation of developing large demands.

Tables 6.1, 6.2, and 6.3 show an example of how the model can be used as a decision aid in deciding on what expansion plan should be adapted Ð one optimized for low, medium, or high growth.

Table 6.1 shows, along the diagonal, the costs associated with correctly predicting the demand scenarios for the low growth, and low growth does in fact take place, e.g. there is no forecast error.

Now what happens if; • A low or medium growth strategy is implemented, and a higher growth scenario develops?

• A medium or high growth strategy is implemented, and a lower growth scenario develops?

Table 6.2 shows the consequences of such forecast error, either in terms of the MW deficit (-) if demands are greater than expected, or the surplus (+) if demands are less than expected, broken down by generation type.

How can the costs of these deficits or surpluses be determined? There are many ways; here is one method.

a) assume the shortage cost is the extra cost involved with having to rush construction of expensive small generation units to make up the deficit, here taken to be the difference in the cost Kwh of meeting unexpected demand from a series of small combustion turbines, which have small construction lead- times, and the lower cost of meeting these same demands with larger, more economic units with much longer construction lead-times.

b) assume the excess capacity cost is the added cost of cancellation of the (now) unnecessary plants started in anticipation of high demands, here taken as 100% of the construction cost of hydro (e.g. construction on these must proceed) and 20% of the construction cost of the thermal plants (i.e. cancellation costs are assumed 20% of the full cost) 31

Table 6.3 shows the results of these assumptions. As the table indicates, in every instance, the cost of building less capacity than is actually needed is less than the corresponding cost of building too much capacity. Thus, if these “over and under” estimation costs are correct, it pays to plan for low or medium growth, but never for high growth. Of course this conclusion would be altered of it was determined that the cost of making up a shortage in capacity was more than the cost of canceling unnecessary plants. Table 6.1: The Impact of Forecast Error Cost Minimizing Capacity Expansion/Utilization Ð Three Scenarios

Actual Growth

Optimal Low Medium High Strategy (2%) (3.8%) (5.7%) Low Growth (2%) $5.8 x 109

Med Growth (3.8%) $11.5 x 109

Hi Growth (5.7%) $23.4 x 109

Table 6.2 Capacity Shortages (-) and Surpluses (+) Resulting from Forecast Error

Actual Growth

Optimal Low Medium High Strategy (2%) (3.8%) (5.7%) Low Growth (2%) 0 -32,500 MW -86,300 MW

Med Growth +23,900 MW Thermal (3.8%) +8600 MW Hydro 0 -53,800 MW 32,500 MW Total Hi Growth +72,450 MW Thermal +48,500 MW Thermal (5.7%) +13,900 MW Hydro +53000 MW Hydro 0 +86,300 MW Total +53,800 MW Total Q1: How do we determine the cost of underestimating demand?

Q2: How do we determine the cost of overestimating demand? 32

Table 6.3 Estimated Total Cost, if Forecast in Error

Actual Growth

Optimal Low Medium High Strategy (2%) (3.8%) (5.7%) Low Growth 5.8 x 109 5.8 x 109 (2%) $5.8 x 109 +7.2 x 109 +19 x 109 13 x 109 $24.8 x 109 Med Growth 5.8 x 109 11.3 x 109 (3.8%) +12.9 x 109 Hydro $11.3 x 109 +11.8 x 109 + 4.8 x 109 Thermal 23.1 x 109 23.5 x 109 Total Hi Growth 5.8 x 109 11.3 x 109 (5.7%) +20.5 x 109 Hydro +7.87 x 109 Hydro $23.4 x 109 + 14.5 x 109 Thermal +9.7 x 109 Thermal 40.8 x 109 Total 29.0 x 109 Total

7 Further Work & Model Implementation. A further twelve month period is being proposed in order to complete the sensitivity analysis of the model and for thorough checking of all the data that is in it. The following pointers indicate what work should be considered in this year ahead. More details are to be found in the new proposal ÒModeling Electricity Trade in Southern Africa, 1999- 2000Ó[13].

• SAPP GPWG is to set up a personal computer, which will run the SAPP LT model. The PC and the solver software are costing a total of less than $6000 for Purdue University in January 1999. Commercial costs will be higher. • SAPP staff are needing time to work with the model on a weekly basis and to consult with SAPP GPWG and the Purdue modeling team. • February 2000: modeling seminars to each national utility of SAPP. These would be similar to the seminars that were conducted when the sensitivity analysis was completed for the short-term model in February 1998. The interests of each national utility will be considered for these. • An input/output windows interface is to be designed to provide an improved user friendliness for data transfer. This facility will allow non-modeling users to run the model without introducing any errors in the coding. Computer science specialists at Purdue are to be employed on this task. • August 1999: It is proposed that there be a second planning meeting with the regional regulators. This is required for planning an electricity trade and 33

regulation course (two weeks long) for the year 2000. It will be hosted by the Zambian Energy Regulation Board.

• There is to be a modeling of broader economic aspects of electricity trade in SAPP. This includes equitable and efficient tariffs, wheeling rates, water charges, environmental aspects, DSM, and the consideration of uncertain demand growth in the development of least cost investment plans. • November 1999: It is proposed also that there be a consultation and training workshop at Purdue in November 1999. This will be primarily to train the SAPP GPWG members and the regional consultants on the windows version of the LT model.

Bibliography

Copies of the SAPP related publications listed below can be obtained from Chandra Allen; Email [email protected], Fax: USA 765-494-2351

[1] F.T.Sparrow, William A. Masters, ÒModeling Electricity Trade in Southern AfricaÓ, Year One proposal, Purdue University, November 1996. [2] F.T.Sparrow, William A. Masters, ÒModeling Electricity Trade in Southern AfricaÓ, Year Two proposal, Purdue University, June 1997. [3] F.T. Sparrow, William A. Masters, Zuwei Yu, Brian H. Bowen, & Peter B. Robinson. “Modeling Electricity Trade in Southern Africa. Year 2 Interim Report”, Purdue University, USA, August 1998. [4] F.T. Sparrow, ÒThe SAPP Long-Term ModelÐLecture NotesÓ, Cape Technikon Workshop, June/July 1998. [5] F.T. Sparrow, W.A. Masters, Z. Yu, B.H. Bowen, P.B. Robinson et al, ÒModeling Electricity Trade in Southern Africa. First Year Report to the Southern African Power Pool, Purdue University, USA, September 1997. Lusaka, Zambia, March 3-4, 1998 [6] F.T. Sparrow, Brian H. Bowen, ÒModeling Electricity Trade in Southern Africa, USER MANUAL (Second Edition) for the Long-Term ModelÓ, Purdue University, USA, January 1999. [7] F.T.Sparrow, Brian H.Bowen, William A.Masters, Z.Yu, ÒElectricity Trade Policies and the Southern African Power PoolÓ, SAPP Regional Meeting, Windhoek, Namibia, February 1997. [8] F.T.Sparrow, Brian H.Bowen, William A Masters, Zuwei Yu, D.G.Nderitu, ÒElectricity Trade Modeling; A SAPP SeminarÓ, SAPP Regional Meeting, Windhoek, Namibia, February 1997. [9] F.T.Sparrow, Brian H.Bowen, ÒReport to USAID/EAGER Trade on the SAPP (Southern African Power Pool) Meeting held at Windhoek, NamibiaÓ, 12,13 February 1997. [10] P.Robinson, F.T.Sparrow, ÒWheeling Charges and Loose Power Pools: North American Experience and its Relevance for the Southern African Power PoolÓ, Sixth Joint Plenary Session of SAPP Sub-Committee Meetings, Harare, Zimbabwe, May13 1997. [11] Zuwei Yu, F.T.Sparrow, Brian H.Bowen, ÒA New Long-Term Hydro Production Scheduling Method for Maximizing the Profit of Hydroelectric SystemsÓ, IEEE Transactions on Power Systems - accepted for publication in 1998, IEEE PES 1997 summer meeting, Paper #97sm-868. [12] Zuwei Yu, ÒHydrothermal Unit Commitment ModelÓ, Departmental paper, Institute for Interdisciplinary Engineering Studies, Purdue University, April 1997 [13] F.T. Sparrow, William A. Masters, ÒModeling Electricity Trade in Southern AfricaÓ, Year Three Proposal, Purdue University, March 1999.