Simulation optimization via simultaneous perturbation stochastic approximation

Stochastic approximation is a simulation optimization technique that has received much attention. Traditional finite difference-based stochastic approximation schemes require a large number of simulations when the number of parameters of interest is large. We apply simultaneous perturbation stochastic approximation (SPSA), which requires only two simulations per gradient estimate, regardless of the number of parameters of interest. We report simulation experiments conducted on a single-server queue, comparing the algorithm with finite differences and with perturbation analysis (PA). We then consider a transportation problem and formulate it as a stochastic optimization problem to which we propose to apply SPSA.