An Optimization Method with a Hybrid Objective Function for Scheduling a Flow Shop Problem

As manufacturing and servicing operations have been increasingly becoming complicated, scheduling which directly effects the manufacturing cost, production quality, level of services is becoming more challenging. By far, various scheduling methods have been developed in order to support decision making in scheduling. It is noted that, however, when a well-tuned scheduling method is applied to different circumstances where scheduling preference of decision makers may be different laborious parameter tuning of the method is required. Such parameter tuning to some extent limits applicability of the scheduling method. To eliminate parameter tuning of scheduling methods, in this study, we proposed a scheduling optimization method with the hybrid objective function. By using small-scale historical schedules evaluated by decision makers, a learning-based objective function is trained to evaluate new schedules. Moreover, to restrain the evaluation errors of the learning-based method which may be caused by the imbalanced training data, the priori objective function is exploited and integrated into the learning-based objective function. The proposed method is validated by a classical scheduling problem, the flow shop scheduling problem with two machines and multiple objectives. Validation results show that, the method outperforms the methods with individual priori objective and learning-based functions. Therefore, by using small-scale evaluation result, the proposed method can capture preference changes of decision makers and avoid laborious parameter tuning of the optimization method.