Expert systems in aggregate production planning

Two expert systems, written in the OPS5 programming language, have been developed to perform the task of aggregate production planning. The first system, known as AGPLAN, is a traditional expert system that develops a plan strictly from the heuristics used by an expert from a local chemical company. The second system, known as AGPLAN-LP, is a coupled expert system, which combines a linear programming model with heuristics to generate a plan.;The justification for this research is based on a production planning model proposed by Bowman, known as the Management Coefficients Model. This method effectively utilized management's past behavior to formulate future production plans using multiple regression. Bowman's model, however, only relied on average past behavior to formulate a plan. Expert system technology improves upon Bowman's model by allowing a more discriminating use of past behavior when generating a production plan.;Although the vast majority of aggregate planning models proposed by the literature are mathematical in nature, several authors indicate that little evidence exists of their widespread use in industry. This research bridges this gap by providing management with a model which improves decision making while utilizing planning procedures that have been developed using past experience. By combining a linear programming model with a manager's heuristics, AGPLAN-LP provides a plan which incorporates both short term profitability factors and managerial policies designed for long term success.;Several statistical experiments were designed to evaluate the performance of AGPLAN and AGPLAN-LP. Plans were evaluated based on inventory cost and an inventory penalty function which penalizes deviations from desired inventory positions specified by the expert planner. The results indicate that both expert systems significantly reduce inventory costs and provide more desirable inventory positions when compared to historical and hypothetical cases. One simulation over a 16 month period indicated that historical inventory costs would have been reduced by 19% and 12% using AGPLAN and AGPLAN-LP respectively. During another experiment, the expert planner chose the production plans of AGPLAN and AGPLAN-LP over his own plans 89% of the time. The performance of these planning models has allowed them to be classified as intelligent decision support systems.