Estimation of buffer size using stochastic approximation methods

Estimation of buffer size requirement in a serial manufacturing system is formulated as a stochastic optimization problem. Two stochastic approximation methods used in conjunction with a single-run optimization procedure are presented. They are the Kiefer-Wolfowitz and Robbins-Monro methods, and they are shown to be superior to the direct search simplex method. The Robbins-Monro method integrated with finite perturbation analysis, has the best convergence rate. The savings in computation time is proportional to the number of parameters to be estimated