COMPARISON OF 2 APPROACHES FOR IMPLEMENTING MULTIRESERVOIR OPERATING POLICIES DERIVED USING STOCHASTIC DYNAMIC-PROGRAMMING

This paper compares two approaches for implementing reservoir operating policies derived using stochastic dynamic Programming (SDP) models. In particular, operating policies for the Shasta-Trinity system in Northern California are generated using SDP algorithms that employ either multilinear or multidimensional piecewise cubic functions to approximate the cost-to-go function. Release decisions in the simulations are then determined by either (1) interpolating in the policy tables or (2) reoptimizing the policy within the simulation, using the cost-to-go function generated by the SDP. The impact on simulated system performance of several discretization and interpolation schemes in the SDP is also evaluated. Reoptimizing the policy when a decision is made within the simulation resulted in better system performance, particularly when severe penalties were incurred for water and power shortages and coarse discretizations were employed in the SDP.