SIMULATION AND CONTROL OF DISTRIBUTED MULTI-AGENT BASED LOGISTICS SYSTEMS
Open Access
- Author:
- Slaugh, Vincent William
- Graduate Program:
- Industrial Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- None
- Committee Members:
- Dr Soundar Kumara, Thesis Advisor/Co-Advisor
Soundar Rajan Tirupatikumara, Thesis Advisor/Co-Advisor - Keywords:
- multi-agent systems
logistics - Abstract:
- This thesis investigates logistics systems in which distributed entities cooperate to serve demands by allocating supplies and transportation resources. In particular, a simulation framework, the Distributed Resource Allocation Planning and Execution Simulation (DRAPES), was designed and implemented as a distributed discrete-event simulation using agents. Using an agent-based approach, the simulation’s purpose is to determine how local decision rules and processes followed by an agent affect the global performance of the system. Agents represent the distributed decision making entities, or units, of the logistics system, and have clearly defined information sensors, communication channels, local knowledge, and actions to change the state of the system. Also, computationally, the agent-based approach helped to design a successful simulation that has potential in the future to work with emulated software components of a real logistics system. Simple rules from two different control methods were tested on the simulation. First, a decentralized approach follows a process of iterative bidding based on demands’ priorities to determine which unit will fulfill which demand. Second, in a centralized approach, the distributed units send reports to a centralized controller, which in turn assembles the information, solves a mixed-integer program, and distributes the results as orders to the rest of the system. Simulation runs were made using scenarios that tested not only each approach’s ability to generate an initial plan but also its ability to re-plan as the system state changed. The centralized control approach was found to consistently fulfill more demands, and achieve a better system utility than the decentralized approach. Because of computational limitations for solving the mixed-integer program, future work is recommended to improve each of these control methods and consider how a hybrid approach could be implemented. This work was designed and implemented to simulate a real-life logistics system. To protect the proprietary aspects of this system, we use generalized and fictitious names to represent the real-world scenarios.