An iterative method to optimize the operation of reservoir systems

Abstract

Optimization of long-term operation of real-world reservoir systems by dynamic programming has been studied. Computer time requirements make it necessary to use iterative methods in the case of multi-reservoir systems. State-incremental dynamic programming provides a reasonable rate of convergence for serial systems when initial policies are suitably chosen. A new method is proposed that combines dynamic programming with successive approximations and state-incremental dynamic programming. Incremental sequential dynamic programming is especially useful for systems of complex configuration which cannot be optimized with other iterative methods due to very slow convergence. Choices of initial policy, state increment, and the number of states to be considered in each iteration are investigated in two case studies.