Scheduling features selection enhanced dispatching decision for dynamic job shop scheduling: a knowledge and data-based approach

Abstract

Production systems for customizable products are characterized by operational dynamics due to complicated job routings and machine failures, resulting in dynamic job shop scheduling problems. Dispatching rules are the most commonly used and effective solution approaches for dynamic job shop scheduling problems in practice, but individual rules can only provide short-sight solutions. Hence, adaptively choosing the right dispatching rule according to production system status of job shops is required. To improve the accuracy and robustness of data-driven dispatching rule decision, this paper first introduces a knowledge and data based dynamic job shop scheduling framework consisting of scheduling examples generation, knowledge-guided scheduling features selection and data-driven dispatching rule decision. Then a novel knowledge-guided estimation of distribution algorithm (KEDA) is proposed for the scheduling features selection to enhance the performance of the data-driven dispatching rule decision approaches, where KEDA adopts three types of knowledge-guided improvement strategies, namely mutual information guided population initialization, evolutionary fitness guided probability model update for offspring generation and feature selection ratio guided variable neighborhood search. Comprehensive experiments based on real-life job shop scenarios demonstrate the feasibility of the proposed approach framework and the superiority of KEDA over competitive algorithms for scheduling features selection problems.