Simulation-optimization using a reinforcement learning approach

The global optimization of complex systems such as industrial systems often necessitates the use of computer simulation. In this paper, we suggest the use of reinforcement learning (RL) algorithms and artificial neural networks for the optimization of simulation models. Several types of variables are taken into account in order to find global optimum values. After a first evaluation through mathematical functions with known optima, the benefits of our approach are illustrated through the example of an inventory control problem frequently found in manufacturing systems. Single-item and multi-item inventory cases are considered. The efficiency of the proposed procedure is compared against a commercial tool.