Spare parts supply chain network modeling based on a novel scale-free network and replenishment path optimization with Q learning

Abstract

The efficiency of inventory management determines the customers’ buying experience, so a supply chain network with a shorter replenishment time is needed. The supply chain network is hoped to be robust to the stock-out of some distributors in the network under random customer demands. At the same time, replenishment path optimization method with the objective of minimizing the replenishment time is required. After reviewing previous work done in the field of supply network topology, scale-free network is proven to be efficient when it was used to model supply network. In addition, multi-agent based collaborative replenishment model is smarter. But, there is rare research on multi-agent based collaborative replenishment in the supply chain modelled by scale-free network. In this study, we proposed a spare parts supply chain network model based on a novel scale-free network. In this network growth process, the connection probability function of connecting new distributor to the existing distributors in the network, is constructed considering the connection number (for an existing distributor, its connection number means the number of other distributors which have collaborative relationship with it) and inventory capacity of the existing distributors and the transit time between new distributor and existing distributors. The connection probability function is built from the standpoints of both new distributor and the existing distributors. Furthermore, different selection policies are discussed in the network growth process to improve the efficiency. Unlike other replenishment path optimization methods, Q learning takes the advantage of interacting with the environment to make a dynamic decision. So, Q learning is selected to optimize the replenishment path in supply chain network. In the experiment, network static and dynamic performance is analyzed using the indicators: degree distribution, clustering coefficient, centrality and response time. Experimental results showed that the replenishment time of supply chain network which are optimized by Q learning is reduced by approximately 40%. So, the shorter replenishment time of the supply chain network is verified.