Heuristic algorithm for scheduling batch and semi-continuous plants with production deadlines, intermediate storage limitations and equipment changeover costs

Abstract

We describe a heuristic algorithm for scheduling a multiproduct, batch or semicontinuous plant. In particular, we address a problem involving intermediate product draw-offs, raw-material feeds to any stage, finite intermediate storage inserted between all stages, and order-deadlines. The strategy is characterized by discretization of the scheduling horizon, generalized way of handling storage and a flexible method of objective function evaluation. Each order is assigned a priority taking into account the due dates, product importance and scheduling preference. Orders are scheduled sequentially according to priority. The algorithm maintains the status of all units and inventories of all materials throughout the planning horizon. To satisfy an order, a production run for the material is initiated as close to the deadline as unit and tank availability constraints permit. Then orders for feeds to the scheduled task are generated, and requirements for each feed are met by scheduling their respective tasks. This procedure is carried out recursively until the feeds are externally supplied raw-materials. Different schedules are generated by changing the order priorities. For each trial, an objective function comprising inventory costs, changeover costs and deadline violation penalties is calculated. This objective function is used as a criterion for choosing the best schedule. The algorithm has been successfully tested on data from an existing multiproduct plant and was found to give significantly better schedules than those manually generated by plant staff. The heuristic solutions are found to be within 23% of an exact lower bound by relaxing the integrality constraints to the scheduling problem posed as an MILP. The goodness of the heuristic was further statistically analyzed by evaluating point and interval estimates for the optimal solution value and calculating the performance measure of the heuristic. This measure was always less than 8% indicating a powerful heuristic.