Parallel dynamic programming for multi-reservoir system optimization

The paper shows how high performance computing can solve the multi-reservoir system optimization problem. A multi-dimensional dynamic programming (DP) model is developed for the four-reservoir problem. Then, the master-slave parallelization strategy is used to parallelize the serial DP algorithm. A high performance parallel computer is then used to solve the four-reservoir system with various numbers of cores (up to 300 cores). The results show that distributed computing effectively shortens the computation time for the algorithm, the speedup further increase as the number of cores increases and parallel efficiency decreases very slowly. Future work should use the distributed computer memory to alleviate DP's large computer memory requirements.