1 The International Journal on Hydropower & Dams (2007) Volume Fourteen, issue 1, 80 - 83 Combined flow prediction and reservoir control system optimizes production at Hoa Binh H. Madsen and J. Høst-Madsen, DHI Water & Environment, Denmark Long Le Ngo, Water Resources University, Vietnam D. Rosbjerg, Technical University of Denmark A combined flow prediction and control system is described for the optimization of multi-purpose reservoir operation. The system integrates a numerical model for simulation of river flow and reservoir operation with an optimization tool. The procedure is applied for optimization of reservoir operation rule curves of the Hoa Binh reservoir in Vietnam, considering hydropower production and downstream flood control, and demonstrates that it could be possible to generate an extra 210 GWh/year on average. Reservoir operation is a complex problem that involves a number of often conflicting objectives, including flood control, hydropower generation, water supply for various users, navigation control and so on. Traditionally, fixed reservoir rule curves have been used for guiding and managing the reservoir operation. These curves specify reservoir releases according to the current reservoir level, hydrological conditions, water demands and time of the year. Established rule curves, however, are often not very efficient for balancing the demands of the different users. Moreover, reservoir operation often includes subjective judgements by the operators. Thus, there is a potential for improving reservoir operating policies, and even small improvements can lead to large benefits. For optimization of reservoir operation, procedures based on coupling simulation models with numerical search methods have been developed. In this paper the MIKE 11 modelling system developed by DHI Water & Environment is adopted for simulating the flow in the river system and reservoir operations. The structure operation module in MIKE 11 allows for the implementation of complex control strategies, whereby reservoirs can be operated by defining a number of different control strategies with various conditions. The use of several control strategies makes it possible to simulate multi-purpose reservoirs, which take into account a large number of objectives. The MIKE 11 modelling system is combined with a numerical optimization tool that is used to optimize different control variables defined for the reservoir operation strategies. The optimization tool includes a general multi-objective optimization framework that searches for the set of non-dominated or Pareto-optimal solutions according to the trade-offs between the various objectives. The simulation-optimization procedure is used in an off-line mode for optimization of reservoir operation rule curves using historical data. Implementation of the optimised rule curves with MIKE 11 then provides a base- line reservoir operation system. This operation system can be further improved in real-time by fine-tuning the reservoir releases using real-time and forecast information. In this case, the MIKE 11 modelling and reservoir control system uses weather forecasts to provide forecasts of reservoir inflows, which are combined with the optimization tool to derive short-term, Pareto-optimal operation strategies. 1. System description The simulation-optimization approach has been applied for optimizing the operation of the Hoa Binh reservoir in Vietnam. Hoa Binh is the largest reservoir in the country, with an active storage of 5.6 x 10 9 m 3 . It is a multi- purpose reservoir providing flood control, hydropower and water supply. The powerplant is equipped with eight turbines, with a maximum unit capacity of 240 MW corresponding to a total power generating capacity of 1920 MW. It produces on average 7800 GWh/year. The Hoa Binh reservoir is on the Da River, which is the largest tributary of the Red River system. The Red River basin is in the northern and north-eastern part of Vietnam and has a total catchment area of 169 000 km 2 , of which 48% is in China and about 1% in Laos. Three major upstream tributaries the Da, Thao and Lo, join and form the Red River delta near Hanoi. The annual average discharge at Hanoi is about 3700 m 3 /s, of which Da River contributes more than 50%. The seasonal variability is significant with about 80% of the annual rainfall 2 occurring in the flood season from May to October. In the analysis, the conflict between flood control and hydropower generation in the flood season is specifically addressed. Fig. 1. MIKE 11 model setup of the lower part of the Red river basin, including the Hoa Binh reservoir. The MIKE 11 modelling system is set up for the lower part of the Red River basin to simulate inflow to the Hoa Binh reservoir and water flow in the downstream part (see Figure 1). To simulate the releases from the Hoa Binh reservoir, operational structures, including bottom sluice gates, spillways and turbines, are specified as control structures in MIKE 11. The control structures are implemented with control strategies which determine how the structures are operated, based on the reservoir level and the water level at a downstream flood control point in Hanoi. The operations consist of specifying the discharge through turbines as well as opening and closing bottom gates and spillways. The control strategies are defined using a list of logical statements according to priorities of the different controls. Opening and closing of the bottom gates and spillways for flood control are described according to flood control rule curves in terms of target water levels in the reservoir and water levels at the flood control point in Hanoi. The discharge through turbines is defined according to hydropower control curves in terms of critical, lower and upper reservoir level curves. 2. Optimization of reservoir rule curves The control variables to be optimized consist of the reservoir water level curves and water level targets at the flood control point at Hanoi. The operation rules are optimized using a two-step procedure. First, the flood control variables are optimized with respect to two objectives: Flood control in terms of the downstream water level; and, Hydropower potential in terms of reservoir level. In the second step, the hydropower control variables are optimized with respect to the hydropower generation in the flood season and the reservoir level at the end of the flood season (used as a surrogate for hydropower generation in the low flow season). For the optimization, selected data from the historical record are used as input to the MIKE 11 model. The main purpose of the optimization is to highlight the trade-offs between the flood control and hydropower objectives. Multi-objective optimization seeks the non-dominated or Pareto-optimal set of solutions with respect to the given objective functions for evaluation of these trade-offs. A set of solutions are identified, where none of the objective functions can be improved without violating one or more of the others. From this curve (denoted the Pareto front) the decision-maker can choose a preferred strategy. One important benefit of using Pareto optimization is that different objective functions measured in different units can be optimized simultaneously without the need to use a common monetary unit, which is often difficult to apply. The results of the optimization of flood control variables are shown in Figure 2. The optimization problem is defined as minimization of the hydropower deficit compared with the maximum hydropower generation capacity (denoted hydropower objective in Figure 2) and minimization of the maximum water level at Hanoi (denoted flood control objective in Figure 2). As expected, a significant trade-off is observed between the two objectives. That is, an improvement in hydropower generation (decrease of hydropower deficit) can only be obtained by an 3 increase in the maximum water level at Hanoi, and vice versa. The figure shows a balanced optimum solution obtained as part of the optimization, which is seen to provide an appropriate balance between the two objectives. In the figure also shows the point corresponding to using the present reservoir regulations. Importantly, the optimization provides Pareto-optimal solutions that are better with respect to both hydropower generation and flood control. Thus, more optimal flood control rules can be implemented that provide an increase in hydropower production in the flood season without violating the basic flood control objective. Present regulation Balanced optimum 1100 1200 1300 1400 1500 1600 1700 650 675 700 725 750 775 Flood control objective Hydropower objective Pareto optimum Fig. 2. Pareto optimal solutions for optimization of flood control variables compared to the present regulations. Also in the case of optimisation of the hydropower control curves a significant trade-off between the two considered objectives is observed. That is, an increase in hydropower production in the flood season can only be obtained by a decrease of the water level at the end of the flood season, and vice versa. The results of this optimization show that Pareto-optimal solutions can be chosen which are better with respect to both objectives compared with the present regulations, that means, more optimal solutions can be chosen to provide increased hydropower production in the flood season as well as larger reservoir level at the end of the flood season (and hence increased hydropower potential in the low flow season). 3. Simulation with optimized rule curves A 20-year historical record was used as input to the MIKE 11 model to simulate reservoir operation using the present regulations and the balanced Pareto-optimal solution. The hydropower generated in the analysed flood seasons using the two operation strategies, is shown in Figure 3. In most flood seasons the balanced optimum solution provides an increase in hydropower production compared with the present regulations. On average an increase of 1.8 percent is obtained, corresponding to 80 GWh/year. 3000 3500 4000 4500 5000 5500 1963 1964 1966 1968 1969 1971 1975 1977 1978 1979 1981 1983 1985 1986 1989 1990 1991 1992 1995 1996 Year Hydropower (million kWh) Present regulations Balanced optimum 4 Fig. 3. Simulated hydropower generation in the flood season using, respectively, the present regulations and the balanced Pareto-optimal rule curves. The simulated reservoir level at the end of flood season using the two operation strategies is shown in Figure 4. In most seasons the balanced optimum solution provides an increase in water level compared with the present regulations. Importantly, the balanced solution provides a substantial increase in water level in the dry years. On average an increase of about 3 m is obtained. The increase in water level at the end of the flood season provides an increased hydropower potential in the low flow season, corresponding to about 130 GWh/year on average. Thus, in total, the balanced optimum solution offers an increased hydropower production of 210 GWh/year on average without compromising flood control. 80 85 90 95 100 105 110 115 120 125 1963 1964 1966 1968 1969 1971 1975 1977 1978 1979 1981 1983 1985 1986 1989 1990 1991 1992 1995 1996 Year Reservoir level (m) Present regulations Balanced optimum Fig. 4. Simulated water level at the end of the flood season using, respectively, the present regulations and the balanced Pareto-optimal rule curves. 4. Real-time optimization Operation of reservoir systems using optimized rule curves will provide a general optimal operation of the system. To further improve the performance, real-time optimization can be adopted, where real-time and forecast information about reservoir levels, reservoir inflows and water demands for various users are used. In this case, the reservoir system is optimised with respect to the short-term operation, using both short-term and long-term objectives. Often there is a conflict between short-term and long-term benefits, and hence the inclusion of long- term objectives in the optimization is important. For the Hoa Binh reservoir a real-time optimization strategy has been implemented. The control variables that include discharge through turbines and opening and closing of bottom gates and spillways are optimized at 6 hour time intervals in a three-day forecast period. Short-term objectives are defined in terms of hydropower production and flood risk at Hanoi in the forecast period. Long-term objectives are implemented by penalizing the deviation of the reservoir level at the end of the forecast period from the target levels defined by the optimized rule curves. Thus, short-term optimizations resulting in operations that provide large deviations in reservoir level compared with the rule curves are penalized. In the Pareto optimization the trade-off between short-term operation objectives and long-term penalizing terms is evaluated. From the Pareto-optimal set the decision-maker can then choose a preferred solution taking other considerations into account. A real-time optimization test is considered in a situation where a large flood is forecasted in the Da River as inflow to the Hoa Binh reservoir (see Figure 5). The forecasted inflow show a peak about 12 hours after time of forecast, followed by a decrease in the remaining forecast period. Thus, in this case a preferred operation strategy is to try to keep as much water in the reservoir as possible, to reduce the downstream flood risk. That is, a Pareto-optimal solution is chosen that allows a larger water level in the reservoir at the end of the forecast period compared with the flood control rule curve to reduce the water level at Hanoi. At the same time an increase in hydropower production during the forecast period is obtained. The results of this preferred solution are shown in Table 1 and can be compared with the results obtained by operating the reservoir according to the optimized rule curves. Real-time optimization provides an increase in hydropower production in the forecast period of 5.5 GWh (4.3 percent) and a decrease in the maximum water level at Hanoi of 0.81 m. 5 0 5000 10000 15000 20000 25000 30000 15/08/96 16/08/96 17/08/96 18/08/96 19/08/96 20/08/96 21/08/96 22/08/96 Inflow [m 3 /s] Da River Thao River Lo River Time of forecast ForecastObserved Fig. 5. Inflow forecasts at the three upstream tributaries used for optimising short-term reservoir operations. Table 1 Comparison of simulation results using the optimised rule curves and the real-time optimal solution. Reservoir level at the end of the period [m] Hydropower generation [GWh] Maximum water level at Hanoi [m] Optimized rule curves Real-time optimal solution 110.7 116.8 127.3 132.8 12.29 11.48 5. Summary and Conclusions A combined flow prediction and control system has been developed for optimization of multi-purpose reservoir operation. The system combines the MIKE 11 modelling system for simulation of river flow and reservoir operation with a numerical optimization tool. The optimization tool includes a general multi-objective framework for estimation of Pareto-optimal solutions. The simulation-optimization procedure has been applied to optimization of the operation of the Hoa Binh reservoir in Vietnam, considering flood control and hydropower generation. A two-step procedure was adopted for optimization of flood control and hydropower rule curves. The results showed that Pareto-optimal solutions can be chosen that are better with respect to both flood control and hydropower generation in the flood season. In addition, the water level at the end of the flood season can be increased with the optimized rule curves, hence providing a larger hydropower potential in the low flow season. By using the rule curves of the balanced optimum solution an increase in hydropower production of about 210 million kWh on average per year is obtained compared with the present regulations. To improve the reservoir operation further, and hence increase the hydropower potential, a real-time optimization system has been developed which is based on real-time and forecast information about reservoir levels, reservoir inflows and water demands. In this case, short-term operation for a three-day forecast period is optimized considering the trade-off between short-term hydropower and flood control objectives and long-term objectives in terms of deviations from the optimised rule curves. The analysis demonstrates that the real-time optimization and control system improves the performance and enhances the flexibility of the reservoir operation compare with a strict application of the rule curves The Authors Henrik Madsen is Senior Research Scientist in the Water resources Department at DHI Water & Environment. He has more than ten years of experience in hydrological modelling, water resources management, extreme 6 value analysis and flood forecasting. He has managed and participated in several national and international research projects on data assimilation, parameter estimation, optimization and uncertainty assessment in hydrodynamic and hydrological modelling. He is responsible for the R&D activities and software development of DHI’s parameter optimization tools tailored towards model calibration and water resources optimization problems. DHI Water & Environment, Agern Allé 5, DK-2970 Hørsholm, Denmark. Long Le Ngo is a lecturer at the Department of Hydrology and Environment, Water Resources University, Hanoi, Vietnam. He obtained his PhD in 2006 at the Technical University of Denmark on optimization of reservoir operation. He has experience in the use of hydrological and hydrodynamic simulation models to investigate a wide range of river basin management issues, including reservoir optimization. He has participated in several projects on integrated water resources management in Vietnam. Department of Hydrology and Environment, Water Resources University, Hanoi, Vietnam. Dan Rosbjerg is Professor in Hydrology and Water Resources at the Institute of Environment & Resources, Technical University of Denmark. He has more than 35 years of research and teaching experience in hydrological modelling and water resources management. From 1978 to 1989 he was Head of the Institute of Hydrodynamics and Water Resources, and in the period 1989-2001 Director of the Groundwater Research Centre. He was President of the water resources systems commission of the International Association of Hydrological Sciences (IAHS) from 2001 to 2005. Since 1986 he has been the editor of the international scientific journal Nordic Hydrology. Institute of Environment & Resources, Technical University of Denmark, Building 115, DK-2800 Kongens Lyngby, Denmark. Jacob Høst-Madsen is Head of the River and Flood Management Department at DHI Water & Environment. He has comprehensive experience in the management of surface water and groundwater resources. Through national and international professional assignments and through his research career he has accumulated a comprehensive knowledge of modelling of surface water and groundwater resources. He has been responsible for DHI's flood forecasting activities in a number of developing countries and is in charge of the development and application of flood forecasting and related hydrological models at DHI. DHI Water & Environment, Agern Allé 5, DK-2970 Hørsholm, Denmark. . International Journal on Hydropower & Dams (2007) Volume Fourteen, issue 1, 80 - 83 Combined flow prediction and reservoir control system optimizes production at Hoa Binh H. Madsen and. applied for optimization of reservoir operation rule curves of the Hoa Binh reservoir in Vietnam, considering hydropower production and downstream flood control, and demonstrates that it could be. Summary and Conclusions A combined flow prediction and control system has been developed for optimization of multi-purpose reservoir operation. The system combines the MIKE 11 modelling system