Evolutionary Pareto optimizers for continuous review stochastic inventory systems

Multi-objective inventory control has been studied for a long time. The trade-off analysis of cycle stock investment and workload, so called the exchange curve concept, possibly dates back to several decades ago. A classical way to such trade-off analysis is to utilize the Lagrangian relaxation technique or interactive method to search for the optimum in a sequence of single objective optimization problems. However, the field of optimization has been changed over the last few decades since the concept of evolutionary computation was introduced. In this paper, a continuous review stochastic inventory system with three objectives about cost and shortage is resolved by evolutionary computation in order to plan for the control policies under backordering and lost sales. Two evolutionary optimizers, multi-objective electromagnetism-like optimization (MOEMO) and multi-objective particle swarm optimization (MOPSO), are employed to well and fast approximate the non-dominated policies in term of lot size and safety stock. Trade-offs are observed in a non-dominated set that no one excels the others in all objectives. Computational results show that the evolutionary Pareto optimizers could generate trade-off solutions potentially ignored by the well-known simultaneous method. Comparisons between the results of backordering and lost sales indicate that decision makers will make more deliberate choices about lot sizing and safety stocking when unsatisfied demand is completely lost.