Actual Benefits from a Hydroelectric Operations Decision Support System

by Charles D. D. Howard, M. ASCE, and Tung Van Do

Cdd Howard Consulting, Ltd.

Since 1986, a decision support system has been used to guide weekly reservoir release decisions at a hydroelectric plant in the coastal mountains of British Columbia , Canada . The decision support system consists of a calibrated hydrologic model, a reservoir operation planning optimization model, and a real-time generation unit loading model. The net value from the decision support system, and the effectiveness of the actual reservoir operation, were quantified by comparing actual hydroelectric generation with the generation that would have been realized without the software.


The Powell dam provides electricity to a nearby pulp and paper complex. The single reservoir with one power station is not a complicated system to model by present standards. But it is one of the few examples of long-term continuous use of what is now called a "Decision Support System ". The software, which is still in use, was developed in 1986 for an 8086 micro computer. The modeling system consists of a daily hydrologic forecasting model, a weekly stochastic LP reservoir optimization, and a real-time turbine loading DP optimization (not described here).

Experience with the forecasting and optimization was used to examine the benefits which have actually been achieved by the software. The benefits from the turbine loading optimization are obvious to the operators, but could not be quantified because the operating log contains insufficient information. The purpose of this paper is to provide information on the benefits from operation of the decision support system This paper is an update of the earlier paper (Van Do and Howard, 1988), which provides additional details on the modeling procedures.

Hydrologic Forecasts

Over a distance of approximately 40 miles, the watershed area of 184 sq. miles (476 sq. km) rises from sea level to over 6000 feet. There are permanent snow and ice fields at the highest elevations. A continuum of microclimates is caused by distance from the heat sink of the ocean, mountain valleys that block weather systems, and elevation.

Hydrologic forecasting is provided by calibrated models that use both the historical deregulated reservoir inflows, and the estimated historical weather over ten elevation bands. Both data sources contain uncertainties, but the large reservoir (580,000 acre-feet) tends to minimize the importance of temporary errors in reservoir inflow forecasts.

Weekly Stochastic Optimization

The operational goal for the 5 unit, 48 MW hydroelectric plant is to minimize the cost of power purchases from the regional grid. This is accomplished by minimizing spill while maximizing head, and by loading the units efficiently. Accurate generation forecasts avoid costly penalties for discrepancies between the monthly contracts for purchases and the actual amounts of the purchases. An initial optimization at a monthly time step considers a three-year time horizon to determine the optimal reservoir levels at the end of a 12-week time horizon. The hierarchial optimization is based on the weekly updates of the current reservoir level and the probabilistic forecast of inflows for the next 12 weeks.

For the second through to the twelfth week, the linear programming model consists of a set of constraints and decision variables for each of seven levels of probability. For the first week, a single set of constraints and decision variables define the deterministic optimal power release.

The objective function maximizes the expected present value of energy generation over the twelve-week time horizon. The output consists of the recommended power flow for the coming week. A supplementary table shows the probabilities of spill, reservoir levels, and power flows for all future weeks.

Methods for Calculating the Benefits

Four types of analysis were used for the evaluation of the 1970-74 operations.

  1. A deterministic optimization with perfect foresight established the maximum possible energy generation. This was assigned the value of 100 percent.
  2. The average annual kw-hours from actual operations, which were governed by rule curves, were evaluated by using the model's coefficients and assumptions. This established the energy that would be obtained without the decision support system.
  3. A week by week simulation of operation was carried out with the decision support system. For each week, starting from the initial condition, the hydrologic model used weather for the previous week to update the watershed state and to provide the probabilistic forecast of inflows. Then, the stochastic optimization was run from the current reservoir level. Next, the recommended release was entered into a spreadsheet. Then the actual weekly inflow was brought in and the end of week reservoir level and energy output were calculated. This process was repeated week by week for each of the five years. This procedure established the kw-hours that would have been realized if the decision support system had been available and if its recommendations had actually been followed.
  4. The same procedure described above in 3 was used, except the average monthly inflows were used in place of the hydrologic forecast. This method tested the value of the optimization analysis when used with a poor hydrologic forecast.

Evaluation of Benefits

The energy produced by operating with perfect foresight, in a deterministic optimization, provides a theoretical basis for measuring the effectiveness of operation. Table 1 shows that for the five years from 1970 to 1974, the actual operation produced 83.4 percent of the theoretically maximum possible average annual energy. The five-year period from 1970 to 1974 included one year with runoff 26 percent below average and four years above average.

Table 1 Theoretical Operational Efficiency, 1970-1974
Deterministic Optimization
100 percent
Actual Operation
83.4 percent
DSS, Decision Support System
95.1 percent
DSS with Average Inflows
92.8 percent

Table 1 shows that if the Decision support system had been available, it would have been possible to realize 95.1 percent of the theoretical maximum. Thus, the benefit from the decision support system might have been about 12 percent, (95.1 - 83.4), of the theoretically maximum amount of energy available from this watershed and dam. Without the hydrologic forecast model, the decision support system could use average monthly historical inflows to provide 92.8 percent effectiveness, a drop of about 2 percent.

In practice, it is not possible always to operate according to the decision support system recommendations. Table 2 shows the actual experience with the decision support system in operation. It can be seen that in 1989, the maximum possible energy was obtained from this water resource. Since 1989, the actual operation has resulted in an average of about 4 percent less than the maximum. This is the "actual operational effectiveness", which is a measure of how closely the water has been managed compared to what was actually possible with the available information.

Table 2 Actual Operational Effectiveness, Actual/DSS
Effectiveness in Percent


The following conclusions were reached:

  1. The rule curve based, actual operation produced 83.4 percent of the theoretically maximum energy potential. If the decision support system had been available, 95.1 percent of the theoretically maximum attainable energy might have been attained. Compared to rule curve operation, this is a potential improvement of 12 percent. In practice, about 8 percent was actually realized.
  2. The actual operation would be improved by increasing the decision frequency from one week to one day. At the fine level of operation provided by the software, the remaining opportunities for improvement are during brief periods when reservoir storage decisions affect the volume of spill.
  3. Of the 12 percent potential benefit from the decision support system, approximately 2 percent stems from the hydrologic forecast model.
  4. The earlier paper used the decision support system to determine the volume which can be economically reserved in storage for contingencies.
  5. These conclusions probably are not applicable universally. For Powell, the results are approximate - it is impossible to predict how the operators actually would have reacted to more or less information. For actual operation during 1970-74 the decision support system did not exist so it is uncertain how closely the decision support system recommendations actually would have been followed. For the 1989-93 period the actual operation was based on the decision support system so it is uncertain how closely the rule curve would have been followed if the decision support system was not available.
  6. The costs for developing the Powell decision support system were repaid quickly by the benefits, probably within one month of the first year of operation.


Van Do, Tung, and Howard, Charles D. D.; Hydropower Stochastic Forecasting and Optimization, proc. ASCE 3rd Water Resources Operations Management Workshop, Fort Collins , Colo, June, 1988, pp 516-526.

Howard, Charles D. D.; Hydroelectric System Operations Optimization, Great Wall Renewable Energy Exhibition, Session 1.2:Hydroelectric Power, 24-27 October, 2006 ---          


Back to Index
Hydrocomp, Inc.