Effective and interpretable dispatching rules for dynamic job shops via guided empirical learning

The emergence of Industry 4.0 is making production systems more flexible and also more dynamic. In these settings, schedules often need to be adapted in real-time by dispatching rules. Although substantial progress was made until the '90s, the performance of these rules is still rather limited. The machine learning literature is developing a variety of methods to improve them. However, the resulting rules are difficult to interpret and do not generalise well for a wide range of settings. This paper is the first major attempt at combining machine learning with domain problem reasoning for scheduling. The idea consists of using the insights obtained with the latter to guide the empirical search of the former. We hypothesise that this guided empirical learning process should result in effective and interpretable dispatching rules that generalise well to different scenarios. We test our approach in the classical dynamic job shop scheduling problem minimising tardiness, one of the most well-studied scheduling problems. The simulation experiments include a wide spectrum of scenarios for the first time, from highly loose to tight due dates and from low utilisation conditions to severely congested shops. Nonetheless, results show that our approach can find new state-of-the-art rules, which significantly outperform the existing literature in the vast majority of settings. Overall, the average improvement over the best combination of benchmark rules is 19%. Moreover, the rules are compact, interpretable, and generalise well to extreme, unseen scenarios. Therefore, we believe that this methodology could be a new paradigm for applying machine learning to dynamic optimisation problems. (c) 2022 Elsevier Ltd. All rights reserved.

Automated summary

The emergence of Industry 4.0 is making production systems more flexible and dynamic, requiring real-time scheduling adaptation. Machine learning methods have been developed to improve scheduling rules, but they often lack interpretability and generalization. This paper proposes a novel approach that combines machine learning with domain problem reasoning to guide the empirical search for effective and interpretable dispatching rules. The experimental results show that the proposed approach outperforms existing literature in various scenarios, indicating its potential as a new paradigm for applying machine learning to dynamic optimization problems.