Discrete Optimization Makespan estimation in batch process industries: A comparison between regression analysis and neural networks

Batch processing is becoming more important in the process industries, because of the increasing product variety and the decreasing demand volumes for individual products. In batch process industries it is difficult to estimate the completion time, or makespan, of a set of jobs, because jobs interact at the shop floor. We assume a situation with hierarchical production control consisting of a planning level and a scheduling level. In this paper we focus on the planning level. We use two different techniques for estimating the makespan of job sets in batch process industries. The first technique estimates the makespan of a job set by developing regression models, the second technique by training neural networks. Both techniques use aggregate information. By using aggregate information the presented techniques are less time consuming in assessing the makespan of a job set compared with methods based on detailed information. Tests on newly generated job sets showed that both techniques are robust for changes in the number of jobs, the average processing time, a more unbalanced workload and for different resource configurations. Finally, the estimation quality of the neural network models appears significantly better than the quality of regression models