Solving a stochastic reservoir management problem with multilag autocorrelated inflows

Abstract

[1] There are many advantages in taking account of multilag autocorrelation of inflows in a reservoir management problem: The flood and water shortage risks diminish, there are fewer spillages, and the generation of hydroplants downstream from the reservoir increases. The disadvantage is that the number of state variables in the optimization problem increases with the number of lags. This paper shows that it is possible to adequately represent the multilag autocorrelation by a single hydrologic variable, the value of which changes from day to day and is equal to the conditional mean of the daily inflow. The paper also shows how to determine the probability distribution of the hydrologic variable for one day as a function of the hydrologic variable and the inflow of the preceding day. This is done for cases where the inflows are represented by a linear autoregressive (AR) model and linear autoregressive-moving-average (ARMA) model. The reservoir management problem is solved with stochastic dynamic programming (SDP). Numerical results are presented. Copyright 2005 by the American Geophysical Union.