Integrating a general Bayesian network with multi-agent simulation to optimize supply chain management

Abstract

In most manufacturing companies, supply chain management (SCM) has been successfully utilized to reduce costs, reduce risk, and improve customer satisfaction. Typical SCM approaches are the cost-focused just-in-time (JIT) approach and the risk-focused just-in-sequence (JIS) approach. Despite some successes, these two SCM approaches have suffered from ad hoc mismanagement and frequent failure due to high costs and uncontrolled risks. This paper proposes a new SCM approach, decentralized SCM, which incorporates multi-core tiers and platform commonization. To prove its validity, we have implemented a multi-agent simulation (MAS) platform named MASSOP (Multi-Agent Simulation-based SCM Optimization Platform) in which a general Bayesian network (GBN) is used to build an intelligent inference mechanism to predict sales and then feed this information back into the MAS engine to optimize SCM by reducing risk and cost. Using a data set compiled over the course of 40 years by an automobile manufacturing company as the basis for preparing the initial parameters for the MAS, MASSOP was built using Repast Simphony and then tested. The results were promising, robust, and statistically valid when compared with JIT and JIS. © 2012 Springer-Verlag.