A Modification Module for Robust Schedule Generation System of Batch Plants.

Abstract

The focus of this paper is to develop a Modification Module (MM) to support the robust schedule generation system in batch process management. Since the tightly coupled nature of schedules in the processes, any local improvement is not effective. Therefore, MM should generate a set of modification tactics to relax the detected multiple bottlenecks simultaneously based not only on the bottleneck information but also on scheduling one. This function is categorized in the pattern recognition problem, thus a backpropagation neural network (BPNN) is a powerful technology applicable to it. In this study, its application is discussed with the emphasis on the following three key aspects. First, to reflect the above tightly coupled nature, global and local features of schedule are examined as the scheduling information of the input layer. Second, due to the processing time variability, the available input data to MM become fuzzy information, and it is based on the property of this scheduling phase. Third, to handle the discretized fuzzy input data, the standard backpropagation learning algorithm is extended. The effectiveness of the proposed MM is demonstrated through simulation experiments on a multiproduct batch process model.