Academic interest in production planning and scheduling problems has surged significantly in the last decade. This is primarily a result of increasing competitive pressure on companies in the chemical process industry to remain profitable in a global marketplace by improving productivity, the efficiency of their production processes, and by minimizing production costs. Concerns about the optimal allocation and usage of resources are addressed by formulating relevant optimization problems, and solving them using appropriate modeling and computational techniques. In the same period, there have been significant developments in computing technology that have yielded affordable computer systems with fast processors, improvements in portable memory capacities, and faster networking speeds allowing for rapid data transfer. These parallel developments in the chemical process and computer industries have made it possible to solve difficult, larger-scale optimization problems that were intractable using older computing systems. These advancements, notwithstanding, there are still problems of interest in process systems research and industry that remain difficult to solve because they require significant computing resources than is readily available on desktop computing systems. Multisite production scheduling is one such problem.;Most current work in literature has focused on solving single-site production sequencing and resource allocation problems over short time horizons ranging from a few days to weeks. These short-term problems are still difficult combinatorial problems. Extending existing optimization methods to multisite problems with long time horizons results in large-scale problems that are much more difficult to solve, requiring long computing times to generate feasible solutions arid, in some cases, cannot be solved using current optimization techniques alone.;In this thesis, we present an alternative optimization approach for solving the multisite production scheduling problem. This approach combines two methods (a) a mathematical formulation for decomposing large-scale optimization models, and (b) an agent-based optimization framework for collaborative problem solving. The decomposition approach utilizes math programming techniques to partition the multisite problem into smaller-scale optimization subproblems. The agent-based strategy combines different rigorous and heuristic algorithms into a collaborative problem solving environment and uses a collection of computers to search and identify good solutions. More importantly, the agent-based system we have developed is able to solve large-scale optimization problems in significantly less time than other existing optimization techniques.;We demonstrate the utility of our combined mathematical decomposition and agent-based optimization system by applying it to a set of representative multisite scheduling problems. These problems are benchmark scheduling problems which are frequently referenced in literature in the context of short-term production scheduling. We evaluate and characterize the agent system performance by comparing its solutions to other existing methods for small- to medium-scale multisite scheduling problems. Finally, we solve examples of large-scale multisite problems some of which are not solvable by alternative optimization methods. We demonstrate the computational efficiency of the agent system in finding good solutions to these problems.
展开▼
