A method to accelerate the computation of successive approximation

Dynamic programming (DP) method has become one of the most popular tools for formulating, analyzing and optimizing a problem with stochastic characteristic. However, time-consuming computation for optimal results emerges while handling the large-size state space. This paper proposes a simple method to speed up the successive approximation for dynamic programming. The single window and double window approaches are presented and a two-queue stochastic model is chosen to test the effectiveness. Compared with original successive approximation, the numerical results show that the observing window methods indeed decrease the computational effort and the double window method is more efficient than the single window. Moreover, we also draw the conclusion that the better-chosen initial values would enhance the performance of DP processing.