Stochastic Long-Term Hydrothermal Optimization for a Multireservoir System

Abstract
The problem of planning the long-term (multiyear) operation of a multireservoir hydrothermal electric power generation system is solved by a sto- chastic dynamic programming (SDP) algorithm using successive approximations. The hydro system model consists of a set of disjoint hydro chains each modeled by an equivalent reservoir and hydroplant. The inflow to the equivalent reservoir in each hydro chain is modeled as an independent log-normal random variable with a time correlation of lag one. The remaining river inflows in the system are modeled as a function of the equivalent reservoir inflows. Thermal unit and load curtailment cost curves are modeled as piecewise linear and convex. The successive approximations algorithm involves the solution of a 2-state stochastic dynamic programming problem for each hydro chain which has as its objective the minimization of the expected discounted production cost, plus the terminal hydro cost, subject to satisfying a number of constraints on the hydro and thermal system and the monthly demand which is represented by a load duration curve. A production-grade computer program has been developed and tested with data for a real system. Numerical results are reported for two study cases with up to eight reservoirs.