A new stochastic control approach to multireservoir operation problems with uncertain forecasts

Abstract

This paper presents a new stochastic control approach (NSCA) for determining the optimal weekly operation policy of multiple hydroplants. This originally involves solving an optimization problem at the beginning of each week to derive the optimal storage trajectory that maximizes the energy production during a study horizon plus the water value stored at the end of the study horizon. Then the derived optimal storage at the end of the upcoming week is used as the target to operate the reservoir. This paper describes the inflow as a forecast-dependent white noise and demonstrates that the optimal target storage at the end of the upcoming week can be equivalently determined by solving a real-time model. The real-time model derives the optimal storage trajectory that converges to the optimal annually cycling storage trajectory (OACST) at the end of a real-time horizon, with the OACST determined by solving an annually cycling model. The numerical examples with one, two, three, and seven reservoirs are studied in detail. For systems of no more than three reservoirs, the NSCA obtains results similar to those obtained with SDP even using a simple inflow forecasting model AR (1). A hypothetical numerical example with 21 reservoirs is also tested. The NSCA is conceptually superior to the other approaches for problems that are computationally intractable due to the number of reservoirs in the system. Copyright 2010 by the American Geophysical Union.