Phenotype Based Surrogate-Assisted Multi-objective Genetic Programming with Brood Recombination for Dynamic Flexible Job Shop Scheduling

• Published in: 2022 IEEE Symposium Series on Computational Intelligence (SSCI) pp. 1218 - 1225
• Main Authors: Zhang, Fangfang, Mei, Yi, Nguyen, Su, Zhang, Mengjie
• Format: Conference Proceeding
• Language: English
• Published: IEEE 04.12.2022
• Subjects: Brood Recombination; Dynamic Flexible Job Shop Scheduling; Genetic programming; Heuristic algorithms; Job shop scheduling; Phenotype; Processor scheduling; Production; Sociology; Surrogate; Training
• DOI: 10.1109/SSCI51031.2022.10022169

Dynamic flexible job shop scheduling (DFJSS) is an important combinatorial optimisation problem with a large number of real-world applications such as component production in manufacturing. Genetic programming (GP), as a hyper-heuristic approach, has been widely used to learn scheduling heuristics for DFJSS. Brood recombination has been shown its effectiveness to improve the performance of GP by generating more offspring and preselecting only promising individuals into the next generation. However, evaluating more individuals requires more computational cost. Phenotype based surrogate models have been successfully used with GP to speed up the evaluation in single-objective dynamic job shop scheduling. However, its effectiveness on multi-objective dynamic job shop scheduling is unknown. To fill this gap, this paper proposes a novel surrogate-assisted multi-objective GP based on the phenotype of GP individuals for DFJSS. Specifically, we first use phenotypic vector to represent the behaviour of GP individuals in DFJSS. Second, K-nearest neighbour based surrogates are built according to the phenotypic characterisations and multiple fitness values of the evaluated individuals. Last, the built surrogate models are used to predict the fitness of newly generated offspring in GP with brood recombination. The results show that with the same training time, the proposed algorithm can achieve significantly better scheduling heuristics than the compared algorithm. The analyses of population diversity, feature importance, and the number of non-dominated individuals have also shown the effectiveness of the proposed algorithm in different aspects.