Comparison of risk-based optimization models for reservoir management

Abstract

Risk minimization in stochastic systems is a challenging problem and this paper compares results of three different techniques in reservoir management. Two-stage stochastic programming (TSP) for maximizing expected benefits is a well-known method, Fletcher and Ponnambalam (FP) and Q-Learning are the two new methods in reservoir management, all of which can include risk minimization in the objective function. The water price uncertainties caused by deregulated markets are considered in addition to random inflows in optimization and simulation is used to compare the results and to develop a risk versus return trade-off curve. One of the contributions of this paper is to consider risk in the Q-Learning algorithm.