Multi-Agent-Based Simulation and Optimization of Production and Transportation

This thesis addresses the integration of software agent technology, simulation and mathematical optimization within the domain of production and transportation. It has been argued that agentbased approaches and mathematical optimization can complement each other in the studied domain. These technologies have often been used separately, but the existing amount of literature concerning how to combine them is rather limited, especially in the domain of production and transportation. This domain is considered complex since; for instance, the decision making is characterized by many decision makers that are influencing each other. Also, problems in the domain are typically large and combinatorial. The transportation of goods has both positive and negative effects on society. A positive effect is the possibility for people to consume products that have been produced at distant locations. Examples of negative effects are: emissions, congestion, accidents, and large costs for infrastructure investments. Increasing competition, experienced by manufacturers and haulers, acts as a motivation for improving the utilization of often limited and expensive production and transportation resources. It is important to maximize the positive effects of transportation while the negative effects are minimized. We present a rather general agent-based simulator (TAPAS) for simulation of production and transportation. By using agent technology, we have been able to simulate the decision making and interaction between decision makers, which is difficult using traditional simulation techniques. We provide a technical description of how TAPAS was modeled, and examples of how it can be used. An optimization model for a real world ``Integrated Production, Inventory, and Distribution Routing Problem’’ (IPIDRP) has been developed. The identified IPIDRP is in the domain of production and transportation problems. For solving and analyzing the problem, we developed a solution method based on the principles of Dantzig- Wolfe decomposition, which was implemented as a multi-agent system inside TAPAS. The purpose is to improve resource utilization and to analyze the potential effects of introducing VMI (Vendor Managed Inventory). Experiments are performed for quantifying the benefits of VMI and for estimating the effects of an agentification of the decomposition approach. Some advantages and disadvantages of an agentification are discussed in this thesis. The work indicates high potentials for integrating agent technology and mathematical optimization. One direction for future work is to use TAPAS as a tool for evaluating the results that are produced by the optimization algorithm. For real world systems, evaluation of optimization results can be expensive and difficult to carry out, and we believe that simulation can be useful for evaluation purposes.