Learning inventory management strategies for commodity supply chains with customer satisfaction

In this paper, we look at a supply chain of commodity goods where customer demand is uncertain and partly based on reputation, and where raw material replenishment is uncertain in both the amount that is available, as well as the price to pay. Successful participation in such supply chains requires a good inventory management strategy. Actors must find a balance between inventory costs and client satisfaction: structurally high inventory costs reduces the profit, but customers that are faced with a depleted supplier will lose confidence and next time purchase from a competitor. This paper presents a model and a simulation environment to learn successful strategies for participation in this type of supply chains. We combine evolutionary algorithms with logistic theories, and use them in a case in a petrochemical setting. We show that software agents are capable of learning basic and more complex strategies, and that complex learned strategies perform better than basic learned strategies.