A random forest-based job shop rescheduling decision model with machine failures

Abstract

Machine failures are the common disturbances in production scheduling, whose appearances are generally random and uncertain. Rescheduling strategies have been proposed to deal with them. However, performances of these rescheduling strategies depend on status of machine failures, and there is no single strategy for every failure status. Hence, how to select the optimal strategy intelligently when a machine failure occurs becomes an important issue. Since the development of artificial intelligence (AI) and machine learning (ML) techniques, intelligent rescheduling has become possible. In this paper, we propose a new rescheduling decision model based on random forest, an effective machine learning method, to learn the optimal rescheduling strategy in different machine failures. We adopt a genetic algorithm (GA) to generate an initial scheduling scheme. Then we design simulation experiments to obtain data of different machine failures which could influence the initial scheme. In each machine failure, all rescheduling strategies are executed respectively and their performances are evaluated based on delay and deviation, then the best strategy is selected as a label. The random forest is trained based on these data samples with labels. Thus the internal mechanism between machine failures and rescheduling strategies can be learned. We conduct experiments to verify the effectiveness of this proposed method and the results show that accuracy can be as high as 97%. Moreover, compared with decision tree (DT) and support vector machine (SVM), the proposed method illustrates the best performance.