In the emerging dynamic and customer-demand driven economy, there is a need for adaptive planning systems that allow companies in supply chains to plan collaboratively and efficiently with incomplete information, and engage in multi-attribute distributed decision-making to meet changing customer needs and production requirements.; The objective of this research is to develop a decision-making framework and methodology that helps the companies in a supply chain to collaboratively, and in a distributed manner, negotiate and arrive at a global Pareto-optimal solution. By representing players non-hierarchically, collaboration between players from all "tiers" is achieved. No player has complete knowledge about all the costs, constraints and objectives of others in the system.; Compared to previous work, our research presents analytical optimization models for distributed decision-making with limited information sharing, and is the first work to consider integer variables using decomposition methods for collaborative planning in supply chains. We developed and compared distributed decision-making methods based on: (1) Primal Decomposition: developed and tested "extended LL" optimality cuts by combining Benders' and Laporte & Louveaux's (LL) optimality cuts; (2) Dual Decomposition using Lagrangian Duality and; (3) Hybrid Primal-Dual Decomposition using a combination of column generation and Lagrangian relaxation.; The Distributed Decision-making method was programmed in Java, with each optimization problem (player sub-problems) being solved using XPRESS-MP, and applied to two different large-scale applications in logistics and supply chain management using actual industry data: (1) Capacity expansion and allocation of testers and handlers in the semiconductor testing industry; (2) Logistics network optimization and determination of value of collaboration between ocean and inland carriers.; The results obtained by using the distributed decision-making method were verified to be the global Pareto-optimal solutions, by solving the centralized model and comparing the solution. The distributed decision-making method converges finitely to the same global Pareto-optimal solution as the centralized method without disclosing local information. The hybrid column generation and Lagrangian relaxation method also generates primal feasible solutions on which to base the branching decisions, thereby reducing the computational time. In addition, scenario analysis shows that our distributed decision-making method provides the ability to quickly re-optimize and adaptively plan for changing circumstances.
展开▼