A Cooperative Use of Stochastic Dynamic Programming and Non-Linear Programming for Optimization of Reservoir Operation

Abstract

Stochastic Dynamic Programming (SDP) is a conventional procedure to extract optimal reservoir operating policies. SDP applies exhaustive search based on the total combinations of discrete values of state variables. The accuracy of SDP can be improved by increasing the number of intervals of inflow or reservoir volume domain; however, increasing the inflow intervals is limited and increasing the reservoir volume intervals does not guarantee the regular improvement of results. This study attempts to determine the near optimum discrete reservoir volume values by cooperating SDP and non-linear programming (CSDP) optimization modules, so the acceptable performance is achieved by limited intervals of the state variables. Towards this end, the objective function is utilized with a hedging policy to control the amount of the monthly reservoir release. The non-linear programming optimization module embodied in SDP, determines the proper values of reservoir volume taking into account the hedging policy and available constrains. Considering various numbers of inflow and reservoir volume intervals, the conventional SDP results are implemented and compared to CSDP approach. Our comparisons indicate an improved reliability performance obtained in the CSDP technique. A Distance Based Interpolation Formula (DBIF) is also illustrated and applied to employ optimal operating polices for reservoir operation management. Findings show better performance of DBIF as compared to conventional methods like Multiple Linear Regression (MLR) during long-time simulation of reservoir operation.