Stochastic optimization of a water supply system

Stochastic dynamic programming (DP) is used to find the operating policy with the least expected steady state cost for a water supply system consisting of a reservoir and an alternative source. The set of possible decisions consists of a number of release rules, each expressing release as a function of storage, rather than a number of discrete releases, as in the conventional DP approach. A flexible procedure is developed which permits inflow to be described by piecewise linear probability density functions, and removes the constraint that inflow and release must be multiples of the discrete unit of storage. The techniques are applied to a reservoir‐river system, and through simulation, the results are compared with the solution found by conventional DP.