SCM network with multi-agent reinforcement learning

This paper constructs a novel multi-agent reinforcement learning frame to apply to the supply chain management network. Supply chains are ubiquitous in the manufacturing of many complex products. Recent advances in planning, scheduling, and autonomous agent technologies have sparked an interest, both in academia and in industry, in automating the process. Taking the dependencies of the underlying production techniques into account., the supply chain management is NP-hard problem. Reinforcement learning is successfully applied to problems of combinatorial complexity, and multi-agent RL method is one of the most powerful methodologies to deal with dynamical and unpredictable domains. By surveying efficient multi-agent RL methods, this paper presents a multi-agent RL method suitable to the dynamic supply chains. The RL agents make optimal job scheduling for each coordinator, and the multi-agent coordination method makes it possible for the tier and its supply chain partners to realize the optimal routine in the supply chain network.