Evolutionary algorithm for inventory levels selection in a distribution supply chain

This paper presents a genetic based technique to select the optimal inventory levels set-points in a supply chain (SC) for distribution. The problem to find the right set-points for the nodes of a supply chain is extremely important when the chain is managed by an automatic control. In the paper, genetic algorithms (GA) are used to search the optimal level of set-points assuming a hypothetic future demand fluctuation. Results, in terms of operating costs are evaluated on different demand fluctuations and different levels of demand prediction error. In order to carry out the simulation a SC simulator was developed in Matlab framework based on a discrete time event model recently proposed in literature. The simulator allows evaluating the dynamical behavior of the supply chain under any demand variation. It implements the optimal LQG (linear Quadratic Gaussian) control strategy to control both inventory levels and operating costs. The chain controllers need the set-points that become strategic values for the chain performance. In this paper we compare results obtained considering some optimal set-point levels found by genetic algorithms and other two heuristic approaches for choosing the set-points. The goodness of the proposed approach is assessed by using an appropriate function which represents the cost of the net activity in a given time horizon. Results show the goodness of the optimization and an average independence of the relative improved performance from the prediction error of the assumed demand.