Stochastic Linear Programming Models for Coordinated Multi-Reservoir Operation

다국어 초록 (Multilingual Abstract)

An efficient operating plan for multiple reservoir systems should be developed based upon the effects of both long-term storage and short-term release policies. For finding such an efficient operating plan, we may have to rely on a “model”, which is an abstract idealization of the real problem. Operation planning for multiple reservoir systems is a complex and challenging problem because of inherent uncertainties in inflow forecasts. In this study, to deal with inflow uncertainty, the two-stage and multi-stage stochastic linear programming models are developed for the long-term storage and short-term release planning.
For the long-term storage planning, long-term inflow predictions are required for the planning of reservoir releases for water supply and hydropower generation. However, the practice of using several inflow predictions does not fully capture the important elements of decision-making under uncertainty analysis, such as non-anticipating decisions or the serial correlations imbedded in the inflow. In this study, stochastic linear programming (SLP) was applied to deal with inflow uncertainties. Specifically, an SLP model was developed for the coordinated operation of a multiple reservoir system. The model optimizes the monthly reservoir storage variation and was formulated as a multi-period, two-stage SLP based on the “fan of individual scenarios.” The inflow scenarios were generated by a multivariate periodic model AR(1) that considers both serial and spatial correlations.
For the short-term release policies, the multi-stage and scenario-based, stochastic linear program with a recourse model was suggested incorporating the meteorological weather prediction information for daily, coordinated, multi-reservoir operation planning. Stages were defined as prediction lead-time spans of the weather prediction system. Multi-stage scenarios of the stochastic model were formed taking into consideration the reliability of rainfall prediction for each lead-time span. A rainfall-runoff model based on the rainfall forecast was used to generate the future inflow scenarios. For short-term stage (2 days) scenarios, the regional data assimilation and prediction system (RDAPS) information was employed, and for mid-term stage (more than 2 days) scenarios, precipitation from the global data assimilation and prediction system (GDAPS) was used as an input for the rainfall-runoff model. After the 10th day (3rd stage), the daily historical rainfall data were used following the ensemble streamflow prediction (ESP) procedure.
The expected benefit of the stochastic models was analyzed quantitatively based on value of information measure. The results indicated that the solutions of the stochastic model were much more effective than those of the deterministic model with average inflows, and that this effectiveness was also maintained in real-time operation in the presence of uncertainty. The benefit of applying this stochastic model to the Nakdong River basin in Korea was presented. The study implies the greater effectiveness of using the stochastic model in real-time operation in presence of uncertainty.