A simulation based optimization method for multi-stage production-distribution systems subject to non-analytical constraints

This paper deals with inventory cost optimization of a multi-stage production-distribution system subject to non analytical constraints. These kind of constraints need to be evaluated by simulation. Therefore, the method that we propose addresses problems where both the performance evaluation and the constraints are evaluated via a stochastic, discrete-event simulation. It is based on random search in a neighborhood structure called the most promising area. A special attention is given to optimizing the allocation of the simulation budget. We show that under some assumptions, the algorithm converges to a set of local optimal solutions with probability 1. This approach is applied to cost optimization of a production-distribution system subject to fill rate specifications. Numerical experiments show that this new method can efficiently solve the discrete optimization problem proposed in this paper.