A guided genetic programming with attribute node activation encoding for resource constrained project scheduling problem

Created by W.Langdon from gp-bibliography.bib Revision:1.7686

@Article{CHEN:2023:swevo,

• author = "Haojie Chen and Xinyu Li and Liang Gao",
• title = "A guided genetic programming with attribute node activation encoding for resource constrained project scheduling problem",
• journal = "Swarm and Evolutionary Computation",
• volume = "83",
• pages = "101418",
• year = "2023",
• ISSN = "2210-6502",
• DOI = "doi:10.1016/j.swevo.2023.101418",
• URL = "https://www.sciencedirect.com/science/article/pii/S2210650223001918",
• keywords = "genetic algorithms, genetic programming, Resource constrained project scheduling, Guided search, Attribute Node activation encoding, Priority rule",
• abstract = "The large-scale characteristic and complex logic between activities have made priority rules (PRs) are more favoured in actual project scheduling, resulting in the increasing attention of genetic programming (GP) with automatically generating more effective PRs. However, the limitations of encoding and numerous random search operators in existing GPs not only affect the effectiveness of evolved PRs, but also reduce their interpretability. This paper proposes a novel Hyper-Heuristic based Guided Genetic Programming with Attribute Node Activation Encoding for resource constrained project scheduling problem. Uniquely, the proposed method transforms existing single class feature activation encoding into attribute node activation encoding for independently controlling each attribute node, and develops an attribute importance calculation method based on the frequency of attribute occurrence and activation. Based on the importance of subtrees and attributes, four guided and two random local search operators are designed to obtain more characteristic PRs. In addition, a two-stage evolution framework that automatically switches stages through iteration number is constructed to achieve performance sampling and guided generation of PRs. Based on the PSPLIB benchmark, although with fewer attribute inputs, the proposed method can generate more effective PRs with significantly better results compared to 12 existing PRs and PRs evolved from the two latest GPs in all test subsets",

}

Genetic Programming entries for Haojie Chen Xinyu Li Liang Gao

Citations