Folded Dynamic Programming for Optimal Operation of Multireservoir System

Abstract

Dynamic Programming (DP) is considered as a good technique for optimal reservoir operation due to the sequential decision making and ease in handling non-linear objective functions and constraints. But the application of DP to multireservoir system is not that encouraging due to the problem 'curse of dimensionality'. Incremental DP, discrete differential DP, DP with successive approximation, incremental DP with successive approximation are some of the algorithms evolved to tackle this curse of dimensionality for DP. But in all these cases, it is difficult to choose an initial trial trajectory, to get at an optimal solution and there is no control over the number of iterations required for convergence. In this paper, a new algorithm, Folded DP, is proposed, which overcomes these difficulties. Though it is also an iterative process, no initial trial trajectory is required to start with. So, the number of iterations is independent of any initial condition. The developed algorithm is applied to a hypothetical reservoir system, solved by earlier researchers. Operating policy obtained using the present algorithm has compared well with that of the earlier algorithm.