view f: tle ell POWEL DATA ~~~ TECHNICAL REPORT SUBJECTrrASK (title) Review of the BC Hydro Marginal Cost Model (MCM) SINTEF Energy Research Address: Reception: Telephone: Telefax: 7034 Trondheim NORWAY Sem Scelands ve; 11 +4773597200 +4773597250 http://ww . energy. sintef. no E. No. : CONTRIBUTOR(S) f;e Mo O(JjOSSO CLlENT(S) BC Hydro Power Supply Purchasing NO 939350 675 TRNO. DATE CLIENT'S REF. F4754 1998- John W. Taylor PROJECT NO. lIX134. ELECTRONIC FILE CODE at;:fl I:\DOK\11 \BM\98004023. DOC J 0 Einar V remes , Powel Data AS ISBN NO. RES REPORT TYPE RCH DIRECTOR ( 82- 594- 1318- Nils Flatabfi DIVISION LOCATION Power Generation and Market Sem Srelandsvei 11 f4V SIGN. CLASSIFICATION Restricted COPIES PAGES LOCAL TELEFAX +4773 5972 50 RESULT (summary) The report contains a review of the Marginal Cost Model (MCM) developed by Power Supply Operations of BC Hydro for resource optirnsation. The purpose of this work is to determne whether or not the MCM is an appropriate decision support tool and if the program development and maintenance are in accordance with the best industry practices. The MCM is a special purpose program used as decision support in the planning and operation of the Wiliston Reservoir based on input representing: uncertain reservoir inflow , expected power production from the rest of the system , uncertain load , uncertain export market , available import possibilities and thermal power production. The program outputs are: incremental water values, power productions, imports , exports and several other results. The program has a high level of detailed modellng of the BC Hydro system and produces results that are useful for decision support in many application areas. The most important are: power trading, Willston Reservoir forecasting, financial forecasting, risk management and system operation planning. The MCM model has an input data representation and a mathematical formulation well in step with what is used in Norway and elsewhere. The processing appears to be correct , and the main conclusion from the review is that the model is a useful tool and therefore a valuable asset for BC Hydro. KEYWORDS SELECTED BY AUTHOR(S) Hydro Operation Planning Decision Support Optirnsation Simulation (I~ Executive Summary The report contains a review of the Marginal Cost Model (MCM) developed by Power Supply Operations of BC Hydro for resource optimization. The marginal cost is just one of the results from the model. The program has been used for 12 years and has been continuously developed to adapt to the changing planning environment. The purpose of this work is to determne whether or not the MCM is an appropriate decision support tool and if the program development and maintenance are in accordance with the best industry practices. The MCM is a special purpose program used as decision support in the planning and operation of the Wiliston Reservoir based on input representing: uncertain reservoir inflow . expected power production from the rest of the system, uncertain load , uncertain export market , available import possibilities and thermal power production. The program outputs are: incremental water values , power productions , imports , exports and several other results. The program has a high level of detailed modellng of the BC Hydro system and produces results that are useful for decision support in many application areas. The most important are: power trading, Willston Reservoir forecasting, ' financial forecasting, risk management and system operation planning. The main conclusions from this evaluation related to the Scope of Work are: 1. Appropriateness of input data Inflow forecasting and modellng are state of the ar Same input data should be used consistently for all BC Hydro models Uncertainty should be modelled- for the Columbia River and the Pacific North West (PNW) market 2. Test of model results Inflow and cost parameters were changed and model response monitored. Prom these tests , the processing within the MCM appears to be adequate. The models providing the input data were not checked. 3. Computation methodology The MCM optimisation par is based on a mathematical formulation that is well in step with what is used in similar programs in Norway and elsewhere. The simulation methodology is good. The number of input scenarios should be increased to improve the usefulness of the output. 4. Dealing with uncertainties The uncertainties of the physical input variables are modelled as well as possible. 1:\DOK\11\BM\98004023. DOC V.! CI Uncertainty in the spot price for the first 18 months should be modelled. This is already implemented in the MCM but the functionality must be activated. 5. Suitabilty, Significance and Usefulness The MCM contains all relevant information concerning the physical and economic operation of the system. It is therefore possible to build further on this model to get more comprehensive tools for power system operation. The MCM is a useful tool for BC Hydro today and wil probably be even more useful in the future with increased competition and trading. '. The MCM gives a number of results that should be used more extensively than today. Currently only the expected (mean) values are used. Probabilstic output of the model should be used for risk management. 6. Compare with other models The mathematical formulation is commonly used in Norway and elsewhere and is ideally suited for the defined problem. modifications The MCM would be strengthened by the changes currently being proposed. 7. Proposed The recommendations for future use of the MCM: BC Hydro should take advantage of the probabilty that this type of tools wil be even more useful in the future with increased competition and trading. The MCM results should be used more extensively than today. Especially the probabilstic output should be util sed to get a picture of the uncertainty of important quantities such as power production and income. The user environment could be improved to provide better access to the model results for the user. A new user interface based on current technology should be implemented. More people should be trained to use the program for sensitivity studies in order to get a better understanding of program output and interaction between different input parameters. This would enhance the confidence in the MCM and increase the understanding of the operational complexity of hydro- thermal systems. Increased use wil also uncover useful extensions and enhancements. Models like the MCM need continuous development to adjust to changing market conditions , to take advantage of better processing techniques , and to better simulate the hydrology. Input data must also be continuously refined. Sufficient resources should be allocated to do this work. 1:\DOK\11\BM\98004023. DOC ....... .......... ....... """""'''''' ...... .............. .......... .......... ............. ........ ........ ............. ....... "" ........... ..... ... ........ ............. ....... ....... .............. .... ......... Ct TABLE OF CONTENT INTRODUCTION.......................... :..................................... 5 OVERVIEW OF SYSTEM AND PLANNING PROCEDURE ...................................... 7 Power production system characteristics .............................................................. 7 Planning procedure............................................................................................... 7 MARGINAL COST MODEL (MCM) ................. """""""'''''''' ........................... ........... 9 Appropriateness of input data ............................................................................. 10 Inflow forecasting for Wiliston Reservoir.......................................... 10 1.1 1.2 Domestic load forecast......................................................................... Generation forecast.................................................. ............................. 12 Inport market......................................... Export market................................................................... .................... 13 Economic parameters...... ............ ........... """"'"'' 14 Quality of input data............................................................................. 14 Test of model results and output......................................................................... 14 Output from MCM ............... ........ ........................................... 15 Case specification.......................... ....................................................... 15 3 . 3.2 Case results........................................................................................... 17 3.3 Discussion of computation methodology..................... ............. "'" ........ ...... ....... 23 3.3. Optimisation algorithm......... ........... """""""" 23 3.3. Simulation procedure............................................................................ 24 Discussion of assumptions ..................... ............................ ...... ............. 24 Adequacy of model in dealing with uncertainties............................................... 25 Suitabilty, significance and usefulness of model............................................... 26 General comments................................................................................ 26 Thermal operation planning ......................................................... ......... 26 Hydro facilty maintenance scheduling ................................................ 27 Electricity trade purchases or sales ....................................................... 27 Wiliston Reservoir elevation forecasting............................................. 28 Financial forecasts................................................................................ 28 Systems operation planning .................................................................. 29 Comparison with similar models ........................................................................ 29 Discussion of proposed modifications.. ................... ........................................... 32 3.3 3.4 5.4 GENERAL COMMENTS ........ ........................... .............. ........ ........ ................ ......... .... 33 CONCLUSIONS............................................................................................................ 35 REFERENCES .......... ........ ................ .... 36 APPENDIX I: USER DESCRIPTION OF USE OF MARGINAL COST MODEL.............. 37 APPENDIX 2: DETERMINISTIC VERSUS STOCHASTIC MODELLING........................ 46 APPENDIX 3: NORMAL RESERVOIR ELEVATION ......................................................... 49 APPENDIX 4: WATER VALUES FROM THE FIRST CASES STUDIES .......................... 1:\DOK\11\BM\98004023. DOC (J INTRODUCTION The report contains an evaluation of the Marginal Cost Model (MCM) developed by the Power Supply Operations deparment of BC Hydro. The program has a history of 12 years and has been continuously developed to adapt to the changing planning environment. The purpose of this work is to determne whether or not the MCM is a useful decision support tool and if the program is developed and maintained in accordance with the best industry practices. The first par of this work was performed during a stay of 8 work days at BC Hydro where discussion of the input data , program performance and use of results , was the main focus. The report preparation has been done in Norway with valuable assistance from Donald Druce for the discussion and interpretation of the program results. The scope can be summarised as: Validate the appropriateness and accuracy of the input data relative to the model output and use. Verify and test model results and output. Examine and comment on the computation methodology and the capabilty and effcacy of the optimisation algorithm , simulation procedure and assumptions. Determne the adequacy of the model in dealing with input uncertainties and sensitivity related to: flow , load , gas supply, energy and capacity supplies from other plants, and the electricity market. Evaluate the suitabilty, significance and usefulness of the model for Power Supply Operations and Corporate business decision makng. Compare the MCM with other models that are used for similar purposes at hydroelectric utilties in other pars of the world. Comment on similarities and differences with respect to the physical system characteristics and modellng issues. Review and comment on the appropriateness of any proposed modification to the model , its inputs , outputs , computations and assumptions. The complete Scope of Work is given in the section TR3 of the Invitation for Proposals with reference PSQ8- 0 11. The approach and amount of work involved on each item is discussed in each subsection. Discussion of information technology standards is not par of the scope but a few comments are given. We have chosen to organise the report as follows: The first chapter gives a brief introduction. 1:\DOK\11\BM\98004023. DOC (I EJR The second chapter briefly summarises the adopted assumptions and planning procedures of BC Hydro. The third chapter discusses the different items as specified in Scope of Work. The section numbering corresponds to the item number in the Scope. We have tried to separate the different items as well as we can , but there might be some overlapping in the report. The fourth chapter discusses general observations regarding use and development of the MCM model. The fifth chapter concludes the findings. 1:\DOK\11\BM\98004023. DOC (I ~P!! OVERVIEW OF SYSTEM AND PLANNING PROCEDURE Power production system characteristics The production system consists of 29 hydroelectric plants, one conventional thermal and two combustion turbine stations. The thermal plants account for about 9 percent of the capacity but accounts for little energy. The hydroelectric plants are grouped into three: Small hydroelectric plants Columbia River system Peace River system Small hydroelectric plants The majority of these have limited storage capacity and their operation is related to the local hydrology rather than the system load. The year-to- year variabilty in the small hydro generation for a given month is usually less that five percent. Columbia River svstem The Columbia River system consists of four plants with a total capacity of 4722 MW (44% of production capacity). The energy produced is about 45% of the system s average yearly energy production. This river system has many operational constraints and despite its size , it does not contribute to the flexibilty of the system operation as much as expected from its SIze. Peace River system The Peace River system consists of two plants with a total capacity of 3112 MW (32% of the production capacity). The energy produced is about 37% of the system s average yearly energy production. The total storage capacity is 40. 000 millon m . The main flexibilty of the operation of the BC Hydro system is connected to this river system. Planning procedure The objective of operation planning is to maximise the net income from the available hydro resources. The hydraulic coupling, the uncertain resource availability, uncertain market conditions and operational constraints make this to a very complicated optimisation problem. Simplifications are necessary to make such a problem computationally feasible. In the procedure of simplification , it is important to take the system characteristics into account. From the general characteristics listed above the following approach has been used: Schedule the resources where the operating flexibilty is low first and then undertake the final coordination procedure. I :\DOK\ 11 \BM\98004023. DOC ~~~ (I Using this approach , Wiliston Reservoir wil be the one that gets the role as slack reservoir and therefore defines the system marginal cost. There is no complex hydraulic coupling along the waterway and both plants on the Peace River have about the same flow capacity and can use the discharged water. The average yearly inflow is about 75 percent of the reservoir storage capacity. The hierarchy applied in the planning has made it possible to prepare a tailor-made model for the long- term operation planing. It is assumed that a general pUrpose program would not be appropriate to account for the special conditions in the system. The BC Hydro system is in this planning structure very nice to model , but it is stil a challenge to come up with a model formulation that takes maximum advantage of the system attributes. The flexibilty in the operation of the Willston Reservoir makes it possible to use a time resolution of one month , which again gives the opportunity to improve the modellng of other items such as importexport capabilities. All these assumptions simpliy the whole planning problem and the crucial question is then if there is anything lost in the simplification process. We have evaluated the MCM model assuming the assumptions to be correct. However, these basic assumptions are to some extent challenged in section 3.3. 3 where we conclude that some kind of interaction between Columbia River and the MCM model should be modelled. I :\DOK\ 11 \BM\98004023. DOC MARGINAL COST MODEL (MCM) The marginal cost model (MCM), which is a resource optimisation model , is formulated as a stochastic dynamic optimisation problem. The margInal cost is just one of the results from the model. The goal function of the model is to maximise the expected net income for the company in the long run using the generation resources and the possible trading options. The net income for a period is given by: The sum of all sales in the export markets multiplied by sales price minus all imports multiplied by import price minus thermal production multiplied by production cost minus unserved energy multiplied by cost of unserved energy The cost of not fulfiling the firm obligations is included in the goal function as shown above. The physical restrictions of the hydraulic system are modelled and included in the optimisation. The problem is dynamic since the reservoirs can be used to store water from one time period to the next. It means that the decisions made in one time period have an impact on the decisions made in the next time step. Knowledge of the export price in one time step may also increase the possibilty to forecast the price for the next time step, compared to a case where the price in the previous time is not known. The problem is stochastic since the future inflow , load and importexport prices are uncertain. If all the input factors were includeej and modelled correctly, the stochastic dynamic optimisation formulation would give the optimal and correct decisions. However , there is a limit to the size of the problems that can be solved with available computer technology. The number of state variables that can modelled gives one of the major limitations. The state variables are given by the variables that couple the time periods (e. g. reservoir level). The number of different uncertain variables that can be modelled gives another major limit. In the MCM model , the Wiliston Lake storage volume , the water supply in the Pacific Northwest (PNW) and the market price the PNW are all modelled as state variables. The MCM uses the weather as a random variable with the following main implications: I Whenever firm load exceeds system resources plus identified import opportunities , the MCM wil cut load in order to balance load and resources. This condition is likely to occur only when the level of Wiliston Reservoir is extremely low and head losses are substantial. The unserved load is priced at 120 mills/kWh, to reflect the expected loss to industry of a curtailment. This functionality in the MCM serves to extend the feasible range of reservoir levels , but at the same time heavily discourages the release of water at such low levels. 1:\DOK\11\BM\98004023. DOC ~~~ It makes it possible to choose a release from Wiliston Reservoir that is conditional on knowing the effect of weather on the domestic load and on the generation to be supplied by the small hydro plants for the current month - since GMSIPCN usually load follow . this adds a sense of realism and variabilty to the operation - weather is assumed to be randomised on a month to month basis Linking the historical small hydro generation to the weather year acknowledges that these plants generally have small reservoirs with inflows primarily due to rainfall runoff, so their generation pattern is influenced more by weather conditions and local operating concerns than by system load or electricity trade opportunities - a realistic contribution from the small hydro plants is added to the MCM without have to spend any CPU time leaving more time for analysing market conditions The MCM model uses mDntWy time resolution and has a planning horizon of six years. Appropriateness of input data is always possible to marginally improve the modellng. Our evaluation of the appropriateness is based on what is practical and possible from our point of view. It Inflow forecasting for Wiliston Reservoir The inflow forecast to all the reservoirs is given by historical observations from 1973 to 1997. The historical values are corrected for snow pack information using a hydrological model from UBC for Wiliston Lake and the Columbia River system. A simpler regression based model is applied for the small hydro plants. The snow pack correction changes the expected spring/summer inflow for the first year and also reduces the uncertainty in the forecast. The historical expected annual inflows and the standard deviation are shown in Table 3. Table 3. Inflow statistics from 1984- 1997 Columbia River system (MCA+REV +KCL+SEV) Peace River system (GMS+PCN) Small h dro lants 21613 4320 30841 17570 17149 5613 2179 21463 7496 14557 3783 941 In order to be able to compare the inflows from year to year we have only used the inflows from 1984 to 1997 since the system has been expanded in the years previous to 1984. If we use the numbers from Table 3. 1 and the fact that these number are based on 14 observations, the expected inflow is with 68 % certainty between the numbers shown in Table 1:\DOK\11\BM\98004023. DOC (I 2. The numbers in the table show that there is considerable uncertainty also regarding the expected inflow to the system. The actual numbers are somewhat smaller than indicated in the table since data from 1973 to 1997 are used in the MCM model. This type of uncertainty wil be reduced as more inflow statistics become available. This last comment is based on the assumption that there are no climatic changes. Columbia River system Peace River s stem Small h dro lants (20458 22767) (16566 17731) (5361 5864) +/- 5.3 +/- 3.4 +/- 4. In our experience the inflow forecasting and modellng are in accordance with state of the industry practice. However , we have not studied the hydrological snow models and the regression model. Druce has published an article (1) where he shows that the UBC models decrease the error in the inflow forecasting. Domestic load forecast In principle the domestic load forecast is a forecast for all the end users sales where the quantity of the future obligations are not known through contracts. The load forecasting is done once a year with a forecasting horizon of 20 years. Only the forecast for the first 6 years is important for the MCM model , since this is the planning horizon of the model. The load forecast is referred to normal temperature , and the variations in load caused by uncertainty in the temperature are included in the MCM model. In order to be able to quantify the uncertainty in the load forecasting, we have compared forecasts for the same year made at different points in time. The forecasts are shown in Table 3. If we use the difference in the load forecast from one year to the next as an indication for the forecasting uncertainty, the maximum change in the forecast for 96/97 is 1.6 % and 2. 6 % for 97/98. We have not compared with measured load since these data are not corrected for weather variations. The load forecast is referred to normal weather. 1:\DOK\11\BM\98004023. DOC (I Table 3. 94/95 95/96 96/97 97/98 98/99 99/00 00/01 01/02 02/03 03/04 Different load forecasts (GWh) made for same periods and different moments in time. 50445. 51888. 53081.0 53593. 54708. 55803. 56885. 49255. 50480. 52472. 53680. 54719. 56056. 57174. 58058. 48853. 50064. 51644. 52590. 53767. 55363. 56614. 57874. 58833. 46736. 48526. 50036. 51758. 53648. 55651.0 57561.0 58481.0 59383. 60496. If we use this limited number of old forecasts as indication of the load forecast uncertainty, we see that uncertainty in the next years forecast is relatively small compared to the uncertainty in the next years inflow. Generation forecast Generation forecasting is performed for all the generation units that are modelled as input to the MCM model. This includes the small hydro plants, the Columbia River plants and Burrard generation. Currently the Columbia River production is calculated using a spreadsheet model. Input to the calculation are Treaty rules and inflow forecast from the UBC model as described in section 1.1. Due to the complexity of the ''freaty rules , only the expected forecast for the Columbia River system is generated. The generation forecasts for the small hydro plants are given by the historical generation for these plants (monthly generation for the period 1973- 1997). Uncertainty in Columbia River generation forecast should be modelled and included into the MCM model. In principle , a simultaneous optimisation of these two reservoirs would have been preferable. However , due to the complexity of the planning problem this wil be very difficult. Other procedures for coupling the model wil be required as for example iteration between individual marginal cost models. Import market The import market consists of two separate markets, the Alberta and the Pacific Northwest (PNW) markets. The prices in these two markets are specified independently. The import price from Alberta at light load hours is given by the coal price and the price at high load hours is linked to the swap prices for natural gas at AECO- C. Line capacities and the number of light and heavy load hours give the maximum import quantity. I :\DOK\ 11 \BM\98004023. DOC (J~ The future coal price is much more certain than the future gas price. Powerex forecasts , for the current year , the quantity and price of energy available from Pacific Northwest. For the subsequent years a typical market pattern is assumed that reflects BP A's fish flush operation. The import market is split from the export market in order to account for the price differences between heavy and light load hours. The difference between import and export disappears , if modellng of separate energy balances for heavy and light load hours are included in the model as discussed in section 3. 7. Wheeling costs and losses are included in the market pnces. Export market The export markets are essentially the same markets as the import markets: the Pacific Northwest and Alberta. The export price in Alberta is connected to the gas price as described in the previous section. The export price in the Pacific Northwest is based on the COB index for the first 18 months and corrected for wheeling and losses , i. e. referred to the border. The export price after the first . 18 months is described by two simple Markov models. The first Markov model describes the water conditions in the Pacific Northwest (PNW). The model can take two values after the first calendar year, the value defines the market size used in the model. After the first 18 months the value of the market size wil also influence on the market price. The second Markov model describes the prices in the PNW market. The model is currently used from the 18 to last month in the planning period. The model can take three values that combined with the first Markov mQ,qel defines six possible prices. The market size for the first calendar year is given based on what Powerex believes the physical tie lines capacities wil be corrected for the quantity already used by commtted contracts. This is a manual process. The price in the Pacific Northwest is the most important market input parameter since the line capacity (3150 MW) is much larger compared to Alberta (600 MW). If separate energy balances for the high load and light load hours are modelled , the same market model can be used for both import and export. We have checked how the prices are modelled in the MCM model, not the quality of the price forecasting methodology. The input to the MCM model should be the best available. The MCM model does not constrain the price forecasting methodology. Uncertainty in the PNW market should be modelled for the whole planning period as discussed further in section 3. I:\DOK\ 11 \BM\98004023. DOC (j Economic parameters The economic parameters used in the MCM model are: Foreign exchange rate Inflation rate Discount rate Foreign exchange rate The exchange rate shows significant moves in the period from January 1992 and up to April 1998. The lowest value is 1.1748 (January 1992) and highest value is 1.4302 (April 1998). There are large fluctuations within this period despite the main increasing trend. For the first four quarters of the study period , the exchange rate is based on the average of the forecasts from Nesbitt Burns , RBC- , Scotia McLeod and Wood Gundy. The rates for the subsequent periods are taken from the 3 Year Plan - Rate assumptions. Nominal discount and inflation rates The nominal discount rate is the sum of the inflation rate and the real discount rate. The real discount rate is the 8% corporate value and the inflation rate is taken from 3 Year Plan - Rate Assumptions. Important impacts of this economic parameter are: The value of possible future income becomes lower. The value of storing water wil less and the MCM wil therefore tend to use more water in the first year. A high discount rate wil reduce the impact of an uncertain future. be The users of the MCM results must be aware of the discount rate used and its impact on the marginal value of the water. 1.7 Quality of input data The quality of the output from the MCM model results are not better than quality of the input data. The input data to the MCM model should be based on the best that is available. The input data could be improved by for example increasing the length of the inflow statistics, applying more resources and models to exportimport price forecasting etc. The different users within BC- Hydro should use the same input data , same price forecast , same load forecast etc. Test of model results and output The MCM model describes a complicated process and it is therefore difficult to make tests where the result can be calculated in advance , i.e manual calculation ofthe results. We have therefore tried to prepare different cases where we know how some of the results (water values) should change when the input is changed. It is not possible to quantify the change due to complex interaction between model constraints , but we know in advance whether the water 1:\DOK\11\BM\98004023. DOC CJ~ values should increase or decrease. This type of testing wil model Output check the consistency of the from MCM Currently two different types of reports are distributed based on the results from the MCM model. Both reports are prepared by Donald Druce and distributed monthly. The first report contains least information and has the largest distribution list. The report contains median for marginal cost for the first 6 months and figures showing 10 , 50 and 90 percentiles for simulated reservoir level and marginal cost of water for the first 3 years. The report also summarises some of the input to the study. The second report is sent to Powerex. The following values are reported in tables: Probability distribution for outflow from Wiliston Lake Probabilty distribution for Elevation in Wiliston Lake Expected discretionary sales and purchases GWh and $ for the next three fiscal years Commtted sales and purchases In addition it is also possible to report: Probability distributions for sales and purchases Production (GWh) and firm load Spil conditions The results have a time resolution of one month and can be relied upon for the first 3 years. Case specification In order to test the model we have defined six different cases. Each case has been prepared and run by Donald Druce. The diffC:.tent cases are specified as follows: 1. Base Case Describes the import, export market and the physical production system by May 1998. The description is identical to what was used in the May report by Donald Druce. Over the next 18 months the export prices are relatively high , as a result of the water supply shortage in the PNW. The expected export prices beyond that time are lower, since the water supply conditions are expected to be higher , on average . In the base case , the model already has the signal to export heavily when the prices are relatively high and is generally doing that. However , it is running up against constraints on market size (reflecting transmission limits) and system generation limitations in the winter. 2. Burrard production The assumptions are identical to the base case except for that Burrard production is cut by 600 GWh in June and July 1998. 2 The expected water conditions after the first 18 months are given by statistics and are assumed to be independent of the current situation. The current water situation is below average. How far ahead this situation influences on the future water conditions are given by the storage capacity and the dynamics of the PNW system. 1:\DOK\11\BM\98004023. DOC Ct ~EJR When Burrard production is reduced the water values should increase independently on the reservoir level. 3. Export price The export price to the PNW is increased by 20 % for the first 18 months compared to the base case. The water values should increase when the export price increases if the tie line capacities are not already fully utilised. If the tie line capacities are already utilised , increased export price wil not change the water values since extra water in the reservoir cannot be sold in the export market. Remember that the water values describe the marginal value of the water, i. e the value of one unit extra of water in the reservoir. 4. Inflow The inflow forecast for May to September 1998 is increased by 3. 7 %. All inflow scenarios are increased by the same amount Increased inflow should result in decreased water values since the probability of future overflow wil increase. The water value is zero (or even negative if flood damage costs are included) if there is 100 % probabilty of overflow. The water values before September 1998 should therefore decrease. The water values after September 1998 should be independent on the increase in inflow since the future inflow and thus also the probabilty of future overflow for a given reservoir level from that time wil be unchanged. 5. Load The load is increased by 1 % for the first three years and 2 % for the rest of the planning period compared to the base case. The water values should increase when the load is increased. Increased load should result in increased water values since increased load wil increase the probability of curtailment in the future and reduce the probabilty of overflow. Both effects contribute to higher water values. 6. Unserved energy The cost of unserved energy is increased by 20 % compared to the base case , this should result in increased water values. Increased cost of unserved energy should increase the water values since the probabilty of curtailment in the future is higher than zero for most of the reservoirs. For the par of the reservoir where the probabilty is almost 100 % in the base case (e. g the lower par of the reservoir in figure 3. ) should the water values increase by 20 %. 7. Discount rate The discount rate is increased from 8 % to 12 %. Increased discount rate should result in decreased water values. The water values describe the marginal value of the water. The water can either be used today or it can be stored and used some time in the future. When the discount rate increases , the future value of the water decreases. The value of current use wil be the same. The sensitivity to the discount rate wil among other factors depend on the storage capacity. 1:\DOK\11\BM\98004023. DOC w.! (I 2.3 Case results General For each case the following values were observed: Water values as a function of reservoir volume for the first eight months Simulated reservoir level for the first three years of the simulation period Simulated production for the first eight months In addition , the water values for the year 2000 are observed for the base case as function of reservoir volume. The cases are such that they should all give a unique increase or decrease in the water values as described in the case description above. Our testing of the model is therefore based on checking how the water values change compared with what we expected. We have checked mainly the water values since these values are the basis for all other model results. The case results also give an indication of the results sensitivity to uncertainty in different input variables. Discontinuous water values The results from the cases show unexpected variation in the water values. The water values should in theory be a continuously decreasing function of increasing reservoir volume if nonlinearities are not included in the optimisation. Two types of discontinuities can be observed: - Near full reservoir as shown in Figure 3. Minor discontinuities in the middle of the reservoir as shown in Figure 3. 2 and 3.3 There were more of these irregularitfes in the first case studies. However , the reason was quickly identified and new calculations were done. The original water value curves are shown in Appendix 4. The irregularities in these figures were connected to some special constraints. The discontinuities near full reservoir are outside the normal operation range of the reservoir and they wil therefore probably not have any significant effect on the model results. The cause of the discontinuities should , however, be found. Use of debugger as discussed in chapter 4 wil make it easier to find the reason. The normal operation range for the reservoir is shown is Appendix 3. The same appendix also shows the connection between reservoir volume and reservoir elevation in Wiliston Lake. The minor discontinuities in the middle of the reservoir have not been explained. We do not believe this to be an indication of a fundamental problem, but an explanation should be found. 1:\DOK\11\BM\98004023. DOC DmLJrn Energy Research 120 100 3500 Figure 3. 8500 13500 18500 23500 Reservoir volume (Mm3) 28500 33500 38500 Water values as a function of reservoir volume for May 1998 for the base case. 160 140 120 :2 100 = 80 2500 7500 12500 17500 22500 27500 32500 37500 42500 Reservoir volume (Mm3) Figure 3. Water values as a function reservoir volume for December 1998 for the base case. I :\DOK\ 11 \BM\98004023. DOC (I~g 160 140 120. f 100 -; 80 2500 7500 12500 17500 22500 27500 32500 37500 Reservoir volume (Mm3) Figure 3. Water values as a function of reservoir volume for December 2000 for the base case. Stochastic modellng of export market The stochastic market model for the PNW is currently used after the first 18 months of the planning period. The stochastic market model is described by the following values: A variable for market price that can take three values (below (1), on (2) and above (3) average price for the given month). These values below and above average market price represent the uncertainty in the future market price and are based on analyses of historical data. The values represent something similar to +/- one standard deviation. This description is not completely correct , but ilustrates the principle of the modellng. A state variable for water conditions in the PNW area. This variable is directly coupled to the maximum export quantity for the given month and can take two values , the second describing the smallest export quantity. The value is also coupled to the expected price after the 18 month. Figure 3.4 shows the water values for different water conditions and price levels in the PNW for December 2000. The values are in percent deviation from the mean of all price and water states. The results are as expected since price state equal 3 and water state equal 1 gives the highest water value for the whole reservoir level. The lowest price combined with the highest water state gives the lowest water values. The figure also shows that the uncertainty in export price that are modelled results in maximum +/- 5 % change in the water values. The figure also gives an indication of how the different price levels and maximum export quantity effects the marginal value of the water. I :\DOK\ 11 \BM\98004023. DOC .._-. -. CI~ Price 1, water 1 -Price 2; water 1 -Price 3; water 1 -Price 1; water 2 Price 2; water 2 --Price 3; water 2 Reservoir volume (Mm3) Figure 3.4 Deviation in water values as function of reservoir level for different price and water states in the PNW. Sensitivity studies Figure 3. 5 and 3. 6 show changes in the water values from the base case for different case studies. The water values should in theory increase for all cases except for the increased inflow case and increased discount rate case for all reservoir levels as described previously. The changes are presented as percentage changes from the absolute water value in the base case given by Figures 3. 1 and 3. 2. It makes it easier to identify the different cases since the absolute change is relatively small for different pars of the reservoir. All the results are as expected , i.e the water values increase and decrease for the correct cases. Figure 3. 6 also shows almost 20 % increase in the water values for the unserved energy case (20 % increase in curtailment cost) near empty reservoir. This is correct , since at this time of the year , the probabilty of future curtailment is almost 100 % near empty reservOIr. The Burrard case gives lower water values than the base case for December 1998. This is caused by the fact the inexpensive production (600 GWh) which is available in June and July in 1998 in the base case instead is available in year 2000. It is therefore correct that the water values for December 1998 is lower than in the base case. Figure 3. 5 shows that both the export case , the load case and the unserved energy case give increased water values for various pars of the reservoir. The increase for the export case is less than expected , but it is explained by the fact that the export capacity is almost fully util sed in the base case. The marginal value of the water wil therefore not increase when the export price is increased. The value of the export wil of course increase. Modellng of flood losses probably causes the variations near full reservoir for the inflow case. 1:\DOK\11\BM\98004023. DOC /"'' (J~ Burrard -Increased load -UnselVsd energy Export price 45'00 -Inflow Discount rate .15 -20 Reservoir volume (Mm3) Figure 3. Changes in water values from base case for May 1998 for the different case studies. Burrard -Increased load Unserved energy -Export price -Inflow go 5 Discount rate 45'00 Reservoir volume (Mm3) Figure 3. Changes in water values for December 1998 for the different case studies. Figure 3. 7 shows the simulated median reservoir level for the different cases. The reservoir levels are very similar for the different cases and it is not easy to separate them in the figure. The cases with increased cost of unserved energy and increased load give the highest spring reservoir level as could be expected. 1:\DOK\11\BM\98004023. DOC ~~~ ;- ' . , "' ''' .. 672 670 666 ! -Base case 'C 664 \\ I . .. ..Burrard " , -h-"Exportprice 662 I-Inflow i -'--Load Unserved energy :g 660 658 656 654 652 Time period Figure 3. Simulated median reservoir level for the different case studies. The case studies have been used to verify the model results and to indicate the sensitivity to different input variables. The results show for example that it might be more important than expected to update the load forecast more frequently since a change of 1- 2 % is important for the water values as shown in Figure 3. 5 and Figure 3. 6. The forecasting error seems to be systematic, which lead to that even a relatively small forecasting error accumulates to a large amount of energy over the planning period. It is impossible to give a general evaluation of the sensitivity to the different input variables since the sensitivity wil depend on the current state. The sensitivity to the export price would for instance be larger if the tie line capacity were not already fully utili sed in the base case. Conclusions The results from the case studies are as expected except for the minor discontinuities as discussed previously. We therefore conclude , based on the results from our case studies, that the model is correctly implemented. All the results seem reasonable. The testing done is not a proof of that the model is correctly implemented. More testing would always make us more certain that everything is correct. It is always possible to test more. We believe, however , that it is more efficient to train more people to run and understand the model than it is to do more testing. If more people run the model , the probabilty of discovering inconsistencies wil increase significantly. 1:\DOK\11\BM\98004023. DOC DmLJrn(p Energy Research 3.3 Discussion of computation methodology Optimisation algorithm The MCM model can simplified be described by the following maximisation problem: J(x(1)) = max (I.:=1 L(x(k), p(k), u(k), k) S(x(N), N) Subject to x(k+l)=x(k) -u(k) + v (k) and constraints to possibl x(k) and u(k) where: J(x (1))x(k) Future net income as function of reservoir level in the first time period Reservoir level in period k Net income in period k as function of exportimport prices, production and reservoir level v(k) p(k) u(k) Inflow in period k Expectation operator , stochastic variables are p(k) and v(k) Exportimport prices Production in period k Time period , month number Number of months in the planning horizon Value of water by end of month N The described maximisation problep1 can be solved by use of stochastic dynamic programming, which is a standard optimisation technique. This method is perfect if the problem can be solved by this approach. The number of state variables that can be included in the model mainly gives the limitations. The number of state variables is in this case given by the number of reservoirs (one) and the number of uncertain variables that are described by Markov models (one for market size and one for market price). One of the advantages with this method is that it wil always give a solution to the problem. For other methods that depend on iteration , convergence is not always guaranteed. We have not checked how the solution algorithm is implemented and can therefore not comment on the implementation and how effcient (meaning how fast it is compared to what is possible) it is. Presently the optimisation takes approximately 1.5 cpu. minute for each year in the planning period which is more than fast enough. Preparation of input data is without doubt much more time consuming than running the model , days or hours compared to minutes. The problem of efficiency may however be more important when the model is improved as described in section 3. I:\DOK\ 11\BM\98004023. DOC 3.3. Simulation procedure A number of historical years put together specify a scenario. The first scenario would , as it is implemented today, be represented by the historical years (1973 to 1978), the second by 1974 to 1979 etc. The historical year gives the inflow and load for each month. To each scenaro corresponding draw from the two Markov models describing the PNW export market (price and quantity limits) are made. There is only one draw for each historical scenario. The simulation results wil be dependent on the draw from the Markov model. The future is simulated for 25 scenaros. The expected values that are output from the model are the mean of the results from the 25 scenarios. In our opinion the simulation methodology is good. In the future we believe that there would be a need for more scenarios, especially connected to possible future use in risk management. The 25 different outcomes currently used , are a too small number to give a credible probabilty distribution. More scenarios can easily be generated combining the historical years differently and makng new draws from the Markov models. The simulation results are dependent on the draw from the Markov model. If more scenarios are generated based on new draws from the Markov model , the simulation results (expected values etc. ) wil be less dependent on the actual draw. Uncertainty in Columbia River production should be included using the same historical years as for the other uncertain variables. It wil probably increase the uncertainty in the simulation results. The simulation takes 20 seconds cpu time for each year in the planning period. Discussion of assumptions The MCM model relies on two impprtant assumptions: The flexible storage capacity in the Columbia River system is small compared to the Wiliston. It is therefore assumed that possible changes in Columbia River production do not influence significantly on the water values in Wiliston Lake. The PNW market can be modelled with sufficient accuracy using a Markov type model. Both assumptions neglect possible interaction between the different physical systems. The PNW system is a hydro- thermal production system with both hydrological and electrical coupling to the BC Hydro system. The Markov type modellng of PNW is only correct if there is no hydrological coupling and if BC Hydro is a price taker in the PNW market. It is not possible to quantify the effect of these assumptions , but significant changes from April to May 1998 in the official results from the MCM model were to some extent explained by changes in the production strategy for the Columbia River. Table 3.4 shows some ' system statistics ' for different pars ofthe BC Hydro system. Kelvin Ketchum supplied the numbers. The values give a simplified description of the physical system , but they indicate that the flexible storage capacity in Columbia River may not be insignificant compared to Wiliston Lake. 1:\DOK\11\BM\98004023. DOC (t~ Both the previous comments imply that some type of interaction between Columbia River and Peace River should be modelled. It should be noted that the operation of the Mica Reservoir must , to some extent , be coordinated with downstream U. S. project operation under the Columbia River Treaty and Non- Treaty Storage Agreement. This linkage reduces the flexibilty of Mica Reservoir storage for BC Hydro purposes. Storage capacity 18.4 10. 570 640 17. 15. \TWh) Maximum production (GWh/week) Inflow (TWh/ ear) 11.5 It is not possible to quantify the error introduced by these assumptions. There are two slightly different approaches to overcome them: Model the PNW and BC system in one large model. Possible solution approaches are discussed briefly in section 3. Expand the MCM model to include both Peace River and Columbia River production. Include PNW market into the MCM as it is done today. The price model for the PNW could be estimated from a physical model of the PNW or historical variations and the forward price. If a physical model is used , possible hydrological connection or correlation could be included using the same historical years as reference in the simulati ons. BC- Hydro is currently developing a new model for the Columbia System. The interaction between the MCM model and this n w model is not decided upon. Adequacy of model in dealing with uncertainties The MCM model explicitly includes modellng of inflow , temperature and market uncertainty, which are included both in optimisation and simulation. The uncertainty in the physical input variables (temperature and inflow) are modelled as well as possible. Two Markov models model the market uncertainty. These models should also be used for the first 18 months if necessary data is available. It might also be possible to increase the number of discrete price levels in the Markov model. In our own model we have found that 7 discrete levels seems to be appropriate. 3 Small hydro in Table 3.4 includes the Kootenay Canal and Seven Mile. 4 Storage capacity is defined here as live storage , equal to the difference between the maximum and the minimum allowable reservoir levels. For the two multi- year reservoirs (Williston and Mica), it is practically impossible to use all of this storage in one year. Furthermore , Columbia River Treaty and, to a smaller extent Peace River Ice constraints limit BC Hydro s flexibility at these two reservoirs. 1:\DOK\11\BM\98004023. DOC (J~ The model results are also dependent on other uncertain input variables as ilustrated by the case studies. The uncertainty in these variables could be included in three different ways: Include stochastic models of these parameters similar to the uncertainty already modelled. Automatic simulation for different values of the uncertain variables (e. g. high , medium and low load forecast). The user specifies and run the sensitivity cases manually. The first method is only possible in theory. The second method is useful if the probability distribution given by the simulation results are used in further analyses such as risk management. This method can however disguise the result' s sensitivity to the different input parameters. We believe that the best practical method is to do manual sensitivity analyses for the uncertain input parameters , which are not already modelled. The MCM model is in principle well adapted to such analyses since it is relatively fast. The user interface and result presentations are , however , based on old computer technology and this makes it in practice impossible for the users of the results to do such analysis themselves. The model can relatively easily be expanded to include simulation of more uncertainty as described above. However , it is probably more beneficial to star using the already available uncertainty from the MCM model in the different applications than it is to simulate for more uncertain variables. Suitabilty, signifcance and usefulness of model General comments The users own explanation of how they currently use the results from the MCM model is described in Appendix 1. The model contains in principle all the relevant information concerning the physical and economical operation of the BC- Hydro system including production , transmission, consumption and importexport. The model can be applied to a wide rage of analysis where the economic consequence for BC Hydro of changes in some of the input variables to model is wanted. CUlTent use of the model as described in Appendix 1 and below , applies only the expected results from the model. The model contains also probability distributions for all the results. We believe this probabilty distribution should be distributed to the users and applied in the decision process. Thermal operation planning The operation of the Burrard thermal plant is handled in the same way as possible imports. There is a time dependent availability of power production where the price is defined by the marginal cost. The expected price and availability of gas give the marginal cost. 1:\DOK\11\BM\98004023. DOC (J ~PEJR The available capacity of the Burrard plant is also defined by other system requirements such as reactive power support , air quality and water temperature. There is nothing in the formulation of the Burrard operation that takes into account the history when the decision is made. If there is no contract on gas that is coupling the decision on the different stages , this is probably an adequate approach. Starstop costs are not relevant with the time resolution used. A state variable for the operation of Burrard would not have been computation ally feasible. The Burrard is modelled in the MCM in a way that is common for thermal units in hydro dominated systems. A full physical model is never possible , but since the MCM takes into account the main characteristics of the Burrard operation , the results from the MCM should be useful for the refinement of the plan for Burrard. Hydro facility maintenance scheduling Maintenance scheduling is required for: Small hydro units Columbia River system Peace River system The whole planning hierarchy is based on the fact that each of the small hydro units does not significantly affect the results of MCM. The planning of the small units are based on local conditions. These units wil see a market price where they wil principally be price takers. For maintenance planning of these plants , the local conditions combined with the marginal cost of water wil define the timing of the maintenance. The maintenance planning of the Columbia River system is more complicated. The power production plans are input to the MCM, and the relative size of the systems cause significant interaction. The marginal water value is important input when decisions about maintenance of the Columbia River are made. However , the use of Treaty and non-Treaty storage wil have impact on the maintenance decisions. It is important that major changes in the output from the Columbia River systems are checked with each new run of the MCM. The consequences of maintenance schedules of the Peace River system must be evaluated by the MCM itself. Major reductions in the power production capabilities may impact the marginal water values. Reduction in transfer capacity due to maintenance or other reasons for derating may be the worst problem for the utilisation of the system. The MCM must then access the impact of reduced market access capability. Electricity trade purchases or sales The MCM model gives the current marginal value of the water. If the current export price is higher , the right decision is to sell as much as possible. The opposite decision is correct if the exportimport price is below the current water value. The model results are well suited to such analysis. I:\DOK\ 11\BM\98004023. DOC ITrnf? Energy Research The MCM model also outputs simulated reservoir levels , exportimport and marginal values of water for different future scenarios. These values are and can be used in decision support for today s trading future contracts. Currently only the expected future values are used by Powerex in their trading decisions as described in Appendix 1. Today the simulated expected net importexport over a period of time (6- 12 months) ahead is used as a target for the net trading over the same period. The expected water values are used to decide which months to buy and sell. An alternative method would be to use the expected simulated water value as a signal for today s trading in the forward market with less consideration to the overall balance. This method is in principle more correct , but it makes it more difficult to control the future energy balance (risk) as long as there is no risk management tools available to provide both price and volume uncertainty. The MCM model contains the connection between physical constraints (tie- line capacities maximum production etc. ), exportimport price uncertainty, load uncertainty, inflow uncertainty, i. e. the connection between price and quantity uncertainty. This is important information that should be incorporated into risk management. Two different approaches can be applied: Enhancement of a MCM type model to also incorporate functionality for risk management. Using the results (price and production scenarios etc. ) from the MCM model as input to other available risk management tools. Without going into detail the first method is the best , but the second method can probably be implemented faster. Both are improvements from present practice. Wiliston Reservoir elevation forecasting The optimisation par of the MCM gives the incremental water values for different reservoir levels and time stages. The simulation par , gives among other results , the reservoir trajectories for different weather and market price scenarios. There wil always be some uncertainty involved these forecasts due to: Adequate selection of weather sequences and appropriate pairing of these with market scenarlOS Non modelled uncertainty of important input variables as for example Columbia River. For normal cases , the uncertainty introduced by such conditions does not have too much impact on the results. For special cases , one may find that the extreme (maximum and minimum) trajectories of the reservoirs are underestimated. Despite this we believe that the MCM is a useful tool for the reservoir elevation forecasting. Financial forecasts A detailed explanation of current use of the model together with wishes for the future are described in Appendix 1. This description contains current use , deficiencies of the current 1:\DOK\11\BM\98004023. DOC (j process and desired situation. For the deficiencies of the current process quite many of these are more related to how things are organised than to the MCM itself. The optimisation par of the MCM calculates the optimal strategy (incremental water values) and the simulation par gives an outcome for different scenarios. Much of the requested information can be found from the basic model output. Total costs can be derived from the proposed production plans and some additional information (fixed costs , sales income from firm load etc. As for the other applications , only the expected values from the MCM model are used. We believe a wider use of all the results would be more appropriate. The MCM gives the distribution for different market and weather conditions and wil therefore provide results that can be used to check the range of possible future income. Systems operation planning The model is definitely a useful tool in the system operation planning. Despite the simpliication in the planning procedures where the results in one evaluation is input to the MCM, use of this model wil be the last step to evaluate the impact of the decisions on the marginal cost values. Change of strategy for the operation of Columbia River plants wil have to be checked with the MCM to estimate the impact on the system operation. Comparison with similar models The MCM is a speciai purpose model developed to solve a specific operation planning problem. There are several similar programs used around the world to solve hydro operation planning problems. These programs would not necessarily be able to solve the same problem as the MCM due to special conditions within the BC Hydro s system. However , use of similar mathematical formulation is a justification for that the MCM is in step with similar programs. This section briefly describes some approaches used to solve the hydro operation planning problem for the purpose of ilustrating techniques in use. Single-reservoir models The MCM model is based on stochastic dynamic programming (SDP). This is a common and frequently used solution method for similar problems also in other countries. The corresponding model used in Norway (2), (4), was originally based on a paper published as early as 1962 (3), and is based on the same type of modellng and solution algorithm as the MCM model. This model has gone through several improvement stages , one of the recent ones being the modellng of the market price using Markov models as a consequence of the market liberalisation. The Norwegian model is principally the same but has also some differences: The model stores and computes the marginal value of water directly (the method is therefore referred to as the water value method). Future income is not computed. Linearization is used for state values between the discrete values. In the MCM model only inflows and decisions that make you reach a pre-calculated value in the next time step is possible. The inflow is pre- processed in order to make it possible to reach these values. 1:\DOK\11\BM\98004023. DOC (J Linearization could make it possible to reduce the number of discrete reservoir levels which again could allow for improved market modelling (i. e. increase the number of discrete price levels). Markov models for price are usually estimated as continuos autoregressive models which are made discrete in the optimisation. The model allows simulation for continuous price senes. A similar model is also developed by the Swedish company Kraftdata. The principle of modellng and optimisation of the single reservoir model is as far as we know identical to our own model. Both the SINTEF Energy Research model (EOPS) and the Swedish model contain simulation models that distribute the production to the different plants in the physical system using a rather complicated set of.rules. We wil not discuss this aspect since the Peace River System is so simple that the simulation is no problem. The simulation aspects of these models are however interesting if more complicated systems are modelled. Generally, methods based on SDP are superior to other solution methods if the basic stochastic dynamic optimisation problem can be described by a limited number of state variables. Multi-reservoir models The problem can generally be formulated as a multi-state stochastic dynamic optimisation problem. It has up to now been impossible to solve such problems in general. In order to solve the problems, different types of simplifications have been made: Simplify the physical description to one or two reservoirs , as in the MCM model. Assume that all the uncertain input parameters (price , inflow and load) are known , often used with shorter planning horizon. Use a combination of optimisatipn and simulation combined with user interference to find the ' optimal' solution , as in our own multi-reservoir model described below. The continuous development of computer technology has now made it possible to solve multi-state stochastic dynamic optimisation problems with fewer simplifications than described above. Some of the mathematical techniques used are the following: Stochastic Dual Dynamic Programmng (SDDP) Scenario Aggregation Determnistic equivalents of the stochastic problem SINTEF Energy Research' s multi-reservoir model (EMPS-model) This model is used for price forecasting, operation planning and investment planning by almost all the major players in the Scandinavian market. The model contains a detailed physical description of the hydraulic systems and includes tie line constraints between different areas in the market. The model is described in (4). Thermal production units are described by their marginal cost, production capacity and probabilty of failure. One of the important results from the model is a price forecast for each defined area. 1:\DOK\11\BM\98004023. DOC The optimisation algorithm is based on a combination of stochastic dynamc programng, linear programmng and heuristic. The user can to some extent influence on the results from the model. This can be an advantage since it is not possible to formally describe all the . players behaviour in the market. Stochastic Dual Dynamic Programming (SDDP) The method is described in (6), (7), (8) and (9) and has much in common with Stochastic Dynamic Programng (SDP). The main advantage is that the method can solve problems that consist of many state variables. The method can therefore in theory be used to optimise a physical system consisting of a number of different hydro plants and reservoirs connected with tie lines with limited capacity. This is the same problem as solved by the EMPS-model the difference being that the SDDP method in theory gives a more formal correct optimisation of the problem. The method has been implemented and tested both by SINTEF Energy Research and CEPEL , Brazil (8) and (9). Based on our own experience , it wil probably take a long time before the method can solve all long and medium-term planning problems. Currently we are implementing the method into our own commercial medium term planning model , which is used for local planning. The SDDP method does in contrast to the SDP method rely on an iteration algorithm and the method is very time consuming compared to SDP approach. Scenario Aggregation The principle of this method is described in (5). Both scenario aggregation and the method using determnistic equivalents assume that a scenario tree can describe the stochastic processes. In practice this wil limit the number oftime periods in the model to 5- , since the number branches in the tree increas. exponentially. Some results from the University in Trondheim indicate that the problems can as well be solved as Deterministic Equivalents using commercial LP solvers as CPLEX. Deterministic Equivalents The stochastic problem is converted to a large determnistic problem that can be solved by standard LP solvers. The number of periods and branches in the scenario tree gives the limitations. If the scenario tree consist of three branches for each period , the number of nodes are given by (3) 8 - 1= 6560. Each branch needs a separate set of decision variables. Reference (10) describes an implementation based in Benders decomposition. The method has much in common with the SDDP method but it is required that the uncertainty can be described along a scenario tree. 1:\DOK\11\BM\98004023. DOC (J~ Discussion of proposed modifications The following modifications of the MCM model are proposed: heavy load hours (HLH) and light load hours (LLH) and model the possibility of load factoring the Peace River generation. 2) Use of physical tie- line capacities to define limits for export and import 3) Modellng of uncertainty in Columbia River production 4) Restore uncertainty to the export price forecast for the US market. 1) Segregation of We believe that these proposed modifications of the model can and should be implemented within the current framework. The first and the second modifications are tied together. Implementation of these extensions wil increase the computation time. However , the increase in computation time should not be a problem, due to rapidly increasing computer capacities. The third modification can also easily be included if uncertain (generation plans corresponding to the historical years 1973- 1997) production plans for Columbia River are available. The fourth extension is already implemented but not used. The uncertainty is not modelled for the first eighteen months due to two different reasons: Lack of market information The misunderstanding that the forward price includes all the useful information It has been diffcult to estimate a credible stochastic model for the price based on available historical data. However , more statistical data wil be available. An alternative could also be to apply physically based forecasting models for the PNW market. The stochastic price model can then be estimated from tQe results of the physical model. Some people assume that the forward price gives all the information there is about the future price for the first eighteen months. The MCM model maximises the long- term income of the water resources. The correct value of the generation is the market price at the time each GWh is produced , i.e. the spot price. The forward price can be seen as the market' s best estimate of future spot price. Appendix 2 shows the value of modellng uncertainty in the future market price , even though there is a known forward price. The example also to some extent ilustrates the principle of stochastic dynamic programming. 1:\DOK\11\BM\98004023. DOC ~~~ ~~~ ,-,;. (J GENERAL COMMENTS The MCM model is implemented and run by only one. person. The model is and wil be a very important tool for BC Hydro. The model should therefore be understood and used by more people than what is the case today. The basic principle of the current model should be possible to grasp also for non-specialists. More people must therefore be trained in order to understand the methodology. More than one person should be able to maintain the program code. Productivity could be improved by changing the programing environment. It is for example not possible to run a debugger with the current tools. Modern programmng tools are available for Fortran on a PC. The future user interface for the MCM type model should be run from a Pc. It makes it easier to use the model and the model results. It is also easier to implement a modern user interface on a Pc. The computation par of the MCM model can run on a Unix server or at the local , both solutions are possible. The present user interface consists of a number of Ascii fies. This makes it relatively easy to implement a modern Windows based user interface to the MCM model. The input to and the output from the model should be stored in a database. More simple model reports on basic input and results can be available for non-specialists. In order to do sensitivity studies , it is important to keep track of the connection between input and output. More people should run the model. The inputs to the model are market statistics and forecasts , and physical values given by line capacities and the production system. The user can have a good understanding of these values wIthout understanding the MCM algorithm. When physical tie line constraints are modelled , the model does not contain any input that requires any knowledge of the algorithm to be understood. If the different users themselves can do sensitivity analyses , the confidence in the model results wil increase. More people wil thereby gain insight into how the complicated hydrothermal system functions. It is impossible to prove that the MCM gives the optimal decision strategy by comparing what happened for instance the last year with what had happened if we had followed another decision strategy (a kind of benchmarking). The other strategy could for example be based on the forward price and expected inflow and temperature. The reason for this is that the model gives a decision strategy that in the long run wil optimise use of available resources. The simpler strategy could by chance be better for just the combination of temperature , inflow and market price that occurred last year. The long run is in this context specified by the time needed to run into the different combinations which are included in the statistics that the strategy is optimal for. The inflow is for instance represented by 25 years. At least 25 new years must pass before the history has been covered. If we combine this with the statistics for market price , we must conclude that the strategy must be followed for many years before it can be proved that one strategy is better than another with absolute certainty. I:\DOK\ 11\BM\98004023.DOC Jr. (I~ An estimate of the profit gained by using the MCM model could be computed using a simulator that simulates the different decision strategies over many years. The results from the MCM model are not better than the quality of the input. Quality checking of the forecasting methodology for different input (e. g. price forecasting), if possible , might be beneficiaL The stochastic dynamic programng approach gives without doubt the optimal decision as long as the input is correct. The problem may be both nonconvex and nonlinear which make it difficult to solve with other methods. 1:\DOK\11\BM\98004023. DOC Ct CONCLUSIONS Our conclusions are based on written documentation , discussions and the results from the test simulations. It is easier to verify that the modellng is correct than to prove that the model is implemented without errors. If the basic assumptions for the model hold as discussed in section 3.3.3 , the modellng and solution algorithm implemented in the MCM model is sound. The model gives several important results , which should be used more extensively than today. The results from the model include net income , reservoir levels , export , import , thermal production , and firm load delivery. Each output can be given for each month in the planning period (maximum 3 years ahead), or as accumulated numbers for periods of time. The model also contains probability distributions for the same outputs for each month or accumulated periods. These probabilty distributions should be used further if it is interesting for the decision maker to for example know that there is 10 % probabilty of the income for the next year to be below a certain number with the current contract portfolio and production strategy. The probabilty distributions are especially interesting in connection to future use in risk management. The modellng should be improved as suggested by Donald Druce and described in section Because the model is a very important tool , more people should be able to run and maintain the model. The user interface should be improved and run on a Pc. It is probably possible to run also the optimisation on the PC. Use of modern software tools as debuggers can simplify development and maintenance of the program code significantly. Methods for integration of the planned Columbia River model with the MCM model must be developed in parallel with the development of the new model. The value of the MCM is much dependent on consistent use for different applications. If it can be agreed upon that the input data is the best available and is adequately represented in the model , tbe strategies and forecasts developed should be used in the different applications. If new information becomes available, the model should be rerun. To make this possible , the MCM must be upgraded with a new user interface and should also get a database connection to make preparation of tailor-made reports easier. A new user interface would make more people able to use the model and thereby more confident with what the results are and the impact of different assumptions. 1:\DOK\11\BM\98004023. DOC ~~~ , ", " , " ,"," ," , pp. REFERENCES ( 1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Druce , DJ., "Forecasting B. C. Hydros operation of Wilis tone Lake - How much uncertainty is enough" , Stochastic and Statistical Methods in Hydrology and Environmental Engineering, Vol. 3 , pp. 63- , 1995. Kluwer Academic Publishers Netherlands. Wangensteen , I., Mo , B. , and Haugstad , A. Hydro Generation Planning in a Deregulated Electricity Market" , Proceedings from "Hydropower Into the Next Century , Barcelona , Spain , June 1995. Sutton , Uk: Aqua-Media International. Lindquist , J. , 1962. " Operation of a hydrothermal electric system: A multistage decision process , AlEE Journal , April 1962. Haugstad , A. System Modeling in a Hydrothermal Electrical System , Presentation September 5 , 1997 , Melhus , Norway. Gjelsvik , A. , and Wallace , S. W: "Methods for stochastic medium- term scheduling in hydro- dominated power systems, EFI TR A4438 , 1996. Rfitting, T. and Gjelsvik , A. Stochastic dual dynamc programng for seasonal scheduling in the Norwegian power system. vol. 16 , 1991 , pp. 199- 147. Gjelsvik , A. et. aI A case of hydro scheduling with stochastic price model" Proceedings of the 3 international conference on hydropower , Trondheim 1997. A. A. Balkema , Rotterdam , Brookfield , 1997. Pereira , M. , and Pinto , L. Multi-stage stochastic optimization applied to energy planning , Mathematical programmng 52 , 1991. Pereira, M. Optimal stochastic operations scheduling of large hydroelectric systems , Electrical Power & Energy Systems, vol. 43 , no.3, July 1989 , pp. 161- 169. Jacobs, J. , et al. " SOCRATES: A system for scheduling hydroelectric generation under uncertainty , Annals of Operations Research , vol. 59 , 1995 99. 113. I :\DOK\ 11 \BM\98004023. DOC CI~ APPENDIX 1: USER DESCRIPTION OF USE OF MARGINAL COST MODEL Information provided by the Marginal Cost Model (MCM) has a wide range of users. These users either utilize details from the MCM itself, or access other reports which contain information which derives from the MCM. The diagram below highlights the range of users of the MCM. BC Hydro Financial Reporting Powerex Power Supply Operations Power Facilities This appendix contains a description of how the results of the MCM currently are used within in BC Hydro. The descriptions are provided by BC Hydro staff. Only minor editorial changes have been made to the text. In the main report , there are some references to this appendix where the current use and future requirements are discussed. 1 Use of the Marldnal Cost Model (MCM) BC Hvdro Financial Reporting Power Supply Business Services is', the main interface between the BC Hydro Finance and the output from the MCM. However , the end users of this information , either directly or indirectly, is quite broad. The diagram on the next page demonstrates the flow of information which is impacted by MCM output. I :\DOK\ 11\BM\98004023. DOC Prepared by Be Hydro , Power Supply Power Supply Business Services BC Hydro Financial Reporting Provincial Government! BC Utilties Commssion 1:\DOK\11\BM\98004023. DOC Prepared by Be Hydro, Power Supply (J~ What information is currently used and for what purpose A) Power Supplv Business Services Three Year Financial Plan / Current Year Forecast The MCM is a key source for projections involving the Cost of Energy expense line component on BC Hydro s Statement of Operations. Power Supply Business Services uses the MCM summary schedule which lists the three year expected generation at Burrard Thermal (in GWh and $) using valley gas , and spot gas. These are the variable components that go into determning the cost of gas for BUITard included in the Cost of Energy line item on the income statement. This same schedule also provides expected energy volumes and dollar values for purchases from the United States , Alberta and Alcan , as well as sales to the United States and Alberta. This information is not directly included in the Three Year Plan or CUITent Year Forecast because Powerex provides a forecast in its place. However , it is used as an important check against the Powerex forecast as the net difference between sales and purchases must be identical for both the MCM and Powerex forecasts otherwise the predicted future Wiliston elevation levels wil not be valid. Power Supply Business Services uses a monthly MCM schedule which presents commtted purchases and BUITard Thermal generation using assumed swing gas (in GWh and $). This information is directly included in the Three Year Plan or Current Year Forecast for Cost of Energy. Ad- hoc Reports Periodically, Power Supply is reque ted to prepare special ad- hoc reports evaluating the financial impact of potential events. Examples include the analysis of the opportunity cost of the sink- holes at Bennett Dam and the financial impact of measures taken to improve the outlook for Wiliston Reservoir levels. The MCM is able to perform this work , but its output comes in the form of net economic cost. Power Supply Business Services takes this information and translates it into financial reporting terms on a fiscal year basis. An example includes where Power Supply would compare the "before and after " effect on the financial forecast of a significant event , such as the sink- holes found at Bennett Dam. Several assumptions may be run over a period of time as different assumptions change. 1:\DOK\11\BM\98004023. DOC Prepared by Be Hydro. Power Supply (I ~ Opportunities for improvements in the current process for financial forecasting The inability of the model to translate economic costs into a financial statement equivalency. For example , the total impact of a change in circumstances could be $20 millon , but it is not easily known whether what the impact is in terms of sales and purchases and in which fiscal years the components of this total would be realized. The model only looks at marginal cost and not total costs. For example , in planning for Burrard Thermal energy costs it is necessary to add to the costs provided by the MCM any other costs which are allocated , i. e. municipal gas tax , demand charge and prepaid transportation costs which are considered to be sunk costs. The Electricity trade sales and purchase numbers are not the final numbers adopted by BC Hydro. These are prepared by Powerex which uses some MCM information and then develops their own forecast. The MCM uses load forecasting volumes for domestic consumption prepared annually. Periodically, adjustments wil be made based on advice from Marketing & Customer Services for major known changes , Le. industrial labour strikes , El Nino. The financial forecast (as prepared by Marketing & Customer Services) uses the same domestic load forecast as its staring point , but makes refinements to total and monthly curving of consumption on an ongoing basis. These changes are not incorporated in the MCM so cost of energy forecastscan t simply use the numbers provided from the MCM chedules. While the MCM is capable of performng scenario analysis and sensitivity analysis this can only be done by Don Druce. Other users of the output are unable to do any " what if' analysis. In addition , due to Don Druce s availability constraints it is often difficult to test for a large number of scenarios on a timely basis. Although the MCM is too complicated to be easily accessible to all potential users, an improved situation would have additional specialists available to run different analysis and provide expertise for a wide range of users. Comparabilty of different scenarios run over a number of months was sometimes questionable as other inputs would change during the period (Le. updated inflow projections). Therefore , different results couldn t always be directly linked to the manipulated variable. MCM is not useful for treasury s needs for cash flow information. For example , the MCM only provides marginal costs for running BUITard and does not include other cash payments related to this generation such as transportation charges for valley gas which are assumed to be sunk costs by the MCM. There is a limited information available as to what assumptions have been used in the MCM. The monthly Marginal Cost Study states assumptions made for inflation , U. exchange rates , nominal discount rates and the forward prices for gas , Alberta and U. electricity. However , many other assumptions are not stated such as unit availabilty, tieline constraints and operations for Peace River ice. I :\DOK\ 11\BM\98004023. DOC Prepared by Be Hydro, Power Supply ~~~ Desired Situation Predicted load inputs used by MCM should be identical to those provided by Marketing & Customer Service s finance deparment (this is an improvement needed in Marketing & Customer Service , not the MCM). The model should be updated on a regular basis using updated load forecasts. Integration of electricity trade information (sales and purchases reported in the MCM and forecasted by Powerex. Additional staffng resources for performng scenario and sensitivity analysis using the MCM. Output to incorporate' all cost of energy components so that the different scenario analysis ' are directly translatable into financial statement form. Alternatively, an accounting program should be developed which combines the output from the MCM with other costs to produce financial statements. B) BC Hydro Corporate Finance BC Hydro Corporate Finance takes the Three Year Plans, forecasts and ad- hoc reports prepared by Power Supply and described in the last section and combines them with information provided by other business units of BC Hydro to develop consolidated Plans forecasts and reports for BC Hydro. The Cost of Energy expense , which is influenced by the MCM, has a significant impact on these consolidated reports. The end users of these Corporate reports may include some or all of the following: BC Hydro Senior Management BC Utilties Commission Crown Corporate Secretariat Provincial Government should also be noted that the Treasury deparment is a user of the Power Supply reports (and , therefore , an indirect user of MCM output) for projecting cash flow expenditures for energy payments. This work is done to assist with BC Hydro s cash flow management. It 1:\DOK\11\BM\98004023. DOC Prepared by Be Hydro, Power Supply Ci~~ Use of the Mandnal Cost Model (MCM) by Powerex Contribution from Mikael Buchko Powerex reviews the imports and exports determned by the MCM and the resulting Wiliston reservoir levels. Powerex refines the monthly importexport projections while keeping the reservoir level the same at the end of the next water year with the goal of optimising profitabilty in imports and exports. The water values provided by the MCM (Rbch) are used as price indicators in a given month. if Rbch is 12 and the market price is 10 Powerex wil not sell energy from the BC Hydro system. Instead , Powerex would buy energy regardless of the monthly expected volume for the given month as determned by the MCM. Powerex would exceed this energy volume value (if physically possible) and then sell the differenceat a later time in order to maintain the end of water year reservoir elevation. For instance , Another use that Powerex has for the MCM is where significant forward sales or purchases Powerex would request that a model run be performed to determne the sensitivities of the MCM results as a consequence of these actions. have been made or are being considered. Contribution from Murray Margolis Powerex uses both the forecast expected purchase and sales volumes by month and the calculated Rbch value when makng sales and purchase decisions. Commtted sales and purchase contracts are included in both Powerex and MCM forecasts and since Powerex forecasts always leave the ending Wiliston Level unchanged the only difference is in the expected sales and purchases by month. This might include simply transferring sales to time periods that we expect wil have the most value and also increasing both purchases and sales to levels that we feel we can obtain, nd are not captured because of the current limitations of the model. Volumes are used primarily for planning purposes and we do not make purchase and sales decisions within the month solely for meeting volume targets. We do however adjust our marketing decision on a continual bases takng into account aggregate sales and purchase volumes to date and forecasts to the end of the fiscal year including any updates we get on domestic load and system energy. BC Hydro Corporate Finance Requirements As described in Section 1. , par B of this Appendix , BC Hydro Corporate Finance has ongoing requirements for financial forecast information. Powerex provides Electricity Trade Sales and a significant part of the energy purchase component of Cost of Energy to the consolidated forecast. As discussed in this section , the MCM impacts the development of the Powerex forecast and , as a consequence , has an influence on significant and volatile aspects of the consolidated forecast. 1:\DOK\11\BM\98004023. DOC Prepared by Be Hydro , Power Supply Ct~ 3 Use of the Marginal Cost Model (MCM) bv Power Supply Operations (PSO) The MCM has a major impact on how Power Supply Operations (PSO) operates BC Hydro . generating facilities. PSO uses the results of the model primarily for: decisions on running BC Hydro s thermal plants (Burrard , Rupert , etc); except in cases where required for capacity or system support these plants wil only be operated in the marginal cost to do so exceeds the marginal cost of energy in Wiliston , Rbch, as determned by MCM. for decisions on imports/exports that are under PSO' s control (e. g. independent power producers within B.C , energy associated with Non- Treaty Storage activity, etc. ). The value of undertakng these transactions are estimated and compared with Rbch . to determne the costlenefit of compressing maintenance outages in the B. C. Hydro system. If the benefit of a shorter maintenance period , as calculated using Rbch exceeds the incremental cost than the maintenance schedule wil be revised. to provide guidance to PQwerex as to the price ceiling for energy purchases and the price floor for sales. This price is Rbch , as determned by the MCM. as the basis for determning the marginal cost of energy in reservoirs other than Wiliston. There is currently no other means for estimating the value of this energy. On occasion, this results in different price signals for energy from different reservoirs. for performance measurement, keeping track of the margin between the system marginal cost and the actual exportimport transaction price. This is done to evaluate how exportimport decisions have added value to the BC Hydro system. to forecast the Wiliston elevation levels for multiple years. This is a very important function due to the sensitivity of business operations in the town of Mackenzie to reservoir levels. There is an ongoing need for elevation forecasts for both the businesses and BC Hydro to plan their operations accordingly. In addition , this information is relied upon by BC Hydro to communicate the expected levels and to explain the anticipated response, if any, which may be required. to determne the prudent over-winter Peace discharge over the ice regime. This is important for maximizing the hydraulic capacity of the Peace River during the winter months and mitigating against the chance of unwanted flooding in the spring. to perform numerious scenario analysis ' to estimate and compare the cost of possible outcomes. Analysis work of nature occurs on a regular basis and can have a significant influence on the ultimate plan of action under- taken by BC Hydro. The analysis can include a range of activities which have included the following: the options for managing the Bennett Dam sink- hole issue the cost of maintaining minimum elevation levels at Wiliston the options for running or not running the non- SCR units at Burrard testing the options of various longer term export contract proposals to provide a number which indicates the percentage probability of a spil occurring at Wiliston due to high reservoir levels. This is a powerful number for easily describing state of the reservoir. This is just one of the determnants that go into making Rbch. I:\DOK\ 11\BM\98004023. DOC Prepared by Be Hydro, Power Supply the (J~~ the MCM identifies the uncertainties of future operations by providing a range of projected reservoir levels , not just the expected level. This is an important system planning input. 1.4 Use of the Mandnal Cost Model (MCM) by Power Facilties Power Facilities does not use to Marginal Cost Study or the relevant Rbch value as determned by the MCM. However , the System Operating Plan which is prepared by PSO and issued to Power Facilities is a major operational planning document. The System Operating Plan provides information on planned generation at all BC Hydro facilities and additional details for major plants which may include projected reservoir elevation levels, discharge rates and inflows. Information from the MCM is used in developing the System Operating Plan. Assumptions used in the MCM and output from its report is incorporated into PSO' s system planning. Therefore , Power Facilities is an indirect user of the MCM. 5 Other Recipients of Marl!inal Cost Model (MCM) Information There are recipients of the MCM Study report which review the information for a variety of purposes, but do not use any of the output for specific decision makng purposes. These users include the following: Energy Plans & Investments. Power Supply In the past , Energy Plans & Investment (EPI) used the long range marginal cost model to provide energy values for the first two years when calculating CONES (Cost of New Energy Sources). However , the MCM is no longer required as more developed market information has become available in recent years; and a forward price curve has been developed. 1:\DOK\11\BM\98004023. DOC Prepared by Be Hydro , Power Supply CI ~ Marketing & Customer Services Marketing & Customer Services (M&CS) reviews theMCM study report to ensure consistency in assumption and inputs with those used for M&CS load forecast. The MCM Rbch value was used by M&CS when developing a Real- Time Pricing (RTP) rate structure. The MCM value was used as a reasonableness check in evaluating various RTP rate structures. 6 Other Potential Users of the Mandnal Cost Model (MCM Risk Management It is probable that there would be many uses for the MCM in the areas of responsibilty Committee which is expected to evolve from the BC Hydro Energy Risk Management Project. While it is difficult to project these uses until the Commttee is formed and its agenda determned there should be some similarities in their requirements as with those already presented in Section A of this Appendix , Power Supply fallng under the Risk Operating Business Services. Performance Measurement (Transfer Pricing) Although currently not used as a transfer pricing measure , Rbch has been used in the past as a means of allocating a purchase and sales price for energy transactions between Power Supply and Powerex. There remain many strong arguments in favour of using the MCM for determning the transfer price. This is true for both Powerex electricity trade transactions and for market based transactions operated by M&CS such as RTP. 1:\DOK\11\BM\98004023. DOC Prepared by Be Hydro , Power Supply (J APPENDIX 2: DETERMINISTIC VERSUS STOCHASTIC MODELLING Determnistic and stochastic approach to evaluation of a load factor contract In Norway, a type of contract called ' brukstidskontrakt' (load factor contract) is quite popular. Each contract is specified with the following values: The energy (GWh) in the contract. The maximum withdraw from the contract for each time period. (GWh/period). The contract period The contract price (mils/kWh) The flexibility of the contract is given by the number of time periods needed to use the total energy compared to the length of contract period , i. e. the load factor. Assume that we want to sell a load factor contract with energy equal to 10 GWh and 10 GWh/period as maximum withdraw. The contract is assumed to be valid for three periods. It means that all the energy can be withdraw in either the first , second or the third period. This contract is equal to a simple hydro production system with 10 GWh storage capacity, 10 GWh/period as maximum production capacity, zero inflow and initial reservoir of 10 GWh. The rest value of the water in the reservoir is zero at the end of the contract period. Assume that the model shown in Figure A.l describes the spot price for the contract period. The model is similar to the model implemented in the MCM model. The price is known to be 10 mills/kWh in the first period and the figure shows the probabilties for prices in the following two time periods. The expected price is 10 mills/kWh for each time period since the price model is symetric. It is reasonable to believe the forward price to be equal to expected spot price, i.e. 10 mills/kWh for the second and the third time period. If we are risk neutral , what is the value of this contract? First we can use the forward price to evaluate the load factor contract. Forward price evaluation The forward price is 10 mils for the second and the third period. Toady s price is also 10 mills/kWh. The value of the energy is therefore independent of which time period the energy is withdrawn and the value must be: 10 mills /kWh* 10 GWh = 100 000 $ I :\DOK\ 11 IBM198004023. DOC (J Market price 14 10 milslk mills/kWh 6 mills/k Time period Figure A.I Markov model for future spot market price. The figure shows for example that price is 14 millslkWh with probabilty 0. 2 in the third time period if the price is 6 millslkWh in second time period Correct value using SDP By using stochastic dynamic progratng similar to what is used in the MCM model , the correct value of the load factor contract is calculated to 108 400 $ (corresponding to a marginal water value of 10. 84 milsIkWh). The correct value of the contract is therefore 8.4 % higher than the value given by the simple forward price evaluation. The optimal strategy specified by marginal values is shown in Table A.l. The algorithm stars at the last (third) time period. I :\DOK\ 11 \BM\98004023. DOC Ct ~EJR Table A.I 14 millslkWh 10 millslkWh 6 millslkWh 10. The reason for the difference is that the flexibility of the reservoir combined with the spot price uncertainty is incorporated into the calculation. If the price is 6 millslkWh in the second period and the energy in the contract is not used , the optimal thing to do is to wait to the last time period (i.e. use the reservoir , see Table A.l in combination with the Markov model). The value of the flexibilty in the load factor contract is reduced in our example if the probabilty of transition from 6 millslkWh in the second period to 14 millslkWh is reduced. If only the forward price is used in the evaluation , the load factor contract has no additional value compared to a forward contract. The example shows that modellng of uncertainty in spot price may give an optimal value which is different from what can be calculated from the forward price. The difference is given by the flexibility of the contract combined with the price uncertainty. More flexibility and/or increased uncertainty add to the error caused by the determnistic (forward) modelling of market price. The load factor contract is just a simple hydro production system , and the same conclusions are therefore also valid for the MCM model. The example shown above is a simple example that ilustrates the principle , and cannot be used to quantify how much modellng of uncertainty wil improve the MCM model. I :\DOK\ 11 \BM\98004023. DOC ,, (J 674 672 670 668 666 664 - 662 660 J 658 .J 656 654 0: 652 650 648 646 644 642 640 2500 5000 7500 10000 12500 15000 17500 20000 22500 25000 27500 30000 32500 35000 37500 40000 Reservoir volume (Mm3) Figure A 3. Reservoir elevation as function of reservoir volume for Willston I :\DOK\ 11\BM\98004023. DOC Lake CI~~ APPENDIX 4: WATER VALUES FROM THE FIRST CASE STUDIES The appendix shows the water values as function of rtservoir volume for the first case studies. The cause of the irregularities shown in these figures was quickly ideritified and the prevailing figures are shown in the report. 100 g: 50 :; 40 3500 6000 8500 11 000 13500 16000 18500 21000 23500 ' 26000 28500 31000 33500 36000 38500 Reservoir volume (Mm3) Figure A4. Water as function of reservoir volume for May 1998 for the base case in the first case studies (corresponding to Figure 3. 1 in the report) 160 140 120 :c 100 m 80 2000 6000 10000 14000 18000 22000 26000 30000 34000 38000 42000 Reservoir volume (Mm3) Figure A4. Water as function of reservoir volume for December 2000 for the base case in the first case studies (corresponding to Figure 3. 3 in the report). 1:\DOK\11\BM\98004023. DOC
© Copyright 2026 Paperzz