Distributed and collaborative logistics planning and replanning under uncertainty: A multiagent-based approach

In a supply chain, a manufacturing company on certain occasions needs to decide the suppliers from whom the raw materials must be obtained, and the manufacturing sites where the final products must be manufactured, to cater to the needs of specific types of customers. This decision-making problem is termed as procurement problem. Under certain assumptions, such as cost and quality of raw materials and final product are independent of the suppliers and the manufacturing sites, the solution of the procurement problem mainly consists of the least cost means of transportation between the suppliers and the manufacturing sites, and between the manufacturing sites and the customer locations. In this thesis, we address how to solve a procurement problem when the manufacturing company seeks freight companies as the means of third party transportation. We define three research objectives. (1) To develop a distributed problem solving model, define the decision making entities and formalize their interactions. (2) To model a transportation task selection and distribution procedure that minimizes the transportation cost to the manufacturing company and increases the payoff to the freight companies, compared to the current practice. (3) To develop a distributed collaborative algorithm to generate reliable vehicle routes under the assumption that there exists an opportunity for the drivers of the vehicles to transfer goods among each other. The problem-solving model is a multiagent computational model, in which each freight company is represented as comprising of a master agent and several slave agents. A master agent of a freight company represents a computational entity which negotiates with the master agents of other freight companies for transportation tasks, and assigns sub-tasks to the drivers of its vehicles. A salve agent is a computational decision making unit on the behalf of a driver of a vehicle that generates robust routes for the driver of that vehicle. The agents are modeled using Belief-Desire-Intention (BDI) logical model and interact using KQML based messages. The negotiation mechanism consists of a discounted bargaining game model with alternating offers. A distributed algorithm called Collaborative Robust Vehicle Route (CRVR) algorithm is presented for reliable and optimal cost estimation of the tasks in real-time.