A Multi-agent Control Architecture for Supply Chains Using a Predictive Pull-Flow Perspective

Abstract

Supply chains are large-scale distribution networks in which multiple types of commodities are present. In this paper, the operations management in supply chains is posed as a tracking control problem. All inventory levels in the network should be kept as close as possible to the desired values over time. The supply chain state is disturbed due to client demand at the end nodes. A multi-agent control architecture to restore all inventory levels over the supply chain is proposed. First the model for the supply chain is broken down into smaller subsystems using a flow decomposition. The operations management for each subsystem will be decided upon by a dedicated control agent. The control agents solve their problems using a pull-flow perspective, starting at the end nodes and then propagating upstream. Adding new components to the supply chain will have as a consequence the inclusion of more control agents. The proposed architecture is easily scalable to large supply chains due to its modular feature. The multi-agent control architecture performance is illustrated using a supply chain composed of four levels (suppliers, consolidation, distribution, end nodes) using different levels of predictions about client demands. With the increase of prediction demand accuracy the proposed control architecture is able to keep the desired inventory level at the end nodes over time, which makes it suitable for use for just in time production strategies.