Simulation-based method for optimizing multi-echelon inventory systems

We propose a novel simulation-based optimization framework for optimizing distribution inventory systems where each facility is operated with the (r, Q) inventory policy. The objective is to minimize the inventory cost while maintaining acceptable service levels quantified by the fill rates. The inventory system is modeled and simulated by an agent-based system, which returns the performance functions. The expectations of these functions are then estimated by the Monte-Carlo method. Then the optimization problem is solved by a cutting plane algorithm. As the black-box functions returned by the Monte-Carlo method contain noises, statistical hypothesis tests are conducted in the iteration. A local optimal solution is obtained if it passes the test on the optimality conditions.