Terminal normalization in genetic programming for dynamic flexible job shop scheduling

Created by W.Langdon from gp-bibliography.bib Revision:1.8396

@Article{Xu:2025:swevo,

• author = "Binzi Xu and Xinyu Cao and Shuwen Zhang and Jiangping Shen and Chaohua Wang and Songhua Wang and Dengchao Huang and Chun Wang and Yi Mei and Yan Wang",
• title = "Terminal normalization in genetic programming for dynamic flexible job shop scheduling",
• journal = "Swarm and Evolutionary Computation",
• year = "2025",
• volume = "96",
• pages = "101970",
• keywords = "genetic algorithms, genetic programming, Dynamic flexible job shop scheduling, Range archive, Normalization",
• ISSN = "2210-6502",
• URL = "https://www.sciencedirect.com/science/article/pii/S2210650225001282",
• DOI = "doi:10.1016/j.swevo.2025.101970",
• abstract = "The dynamic flexible job shop scheduling problem (DFJSSP) is a challenging NP-hard problem that has been studied for decades. Among various optimisation algorithms, genetic programming hyper-heuristic (GPHH) excels in learning dispatching rules (DRs) for online scheduling. However, an overlooked phenomenon in GPHH is that terminals with larger magnitude may overshadow smaller ones, potentially limiting GPHH's effectiveness. The min-max normalization of terminal values may resolve this issue. Thus, this study aims to investigate its impact on GPHH and its potential benefits. Since the terminal ranges used for normalization cannot be obtained directly and in advance within the current GPHH framework, a range archive (RA) enhanced mechanism is proposed for normalization, including three strategies: RA-lite, RA-all, and simulation-RA. These strategies are evaluated on a multi-objective DFJSSP targeting mean tardiness and total energy consumption. Experimental results on small function set demonstrate the presence of inconsistent terminal ranges, in which case normalization can effectively address this issue and enhance the optimisation performance of GPHH. While experimental results on large function set indicate that multiplication and division achieve a similar effect, suggesting limited benefit from normalization. However, further analysis reveals that normalization, particularly simulation-RA, can produce more interpretable DRs without compromising performance",

}

Genetic Programming entries for Binzi Xu Xinyu Cao Shuwen Zhang Jiangping Shen Chaohua Wang Songhua Wang Dengchao Huang Chun Wang Yi Mei Yan Wang

Citations