Simulation-based optimization by combined direction stochastic approximation method

This paper proposes the combined direction stochastic approximation method for solving simulation-based optimization problems. The new algorithm is a stochastic analogy of conjugate gradient method, which employs a weighted combination of the current noisy negative gradient and some former noisy negative gradient as iterative direction. Our numerical experiments show that the new algorithm outperforms the classical RM algorithm for two typical simulation-based optimization problems, a.e., M/M/1 queuing problem and inventory problem.