A learning interactive genetic algorithm based on edge selection encoding for assembly job shop scheduling problem. (September 2021)

• Record Type: Journal Article
• Authors: Guo, Weifei; Lei, Qi; Song, Yuchuan; Lyu, Xiangfei
• Is Part Of: Computers & industrial engineering. Volume 159(2021)
• Publication Date: 2021-09
• Subjects: Genetic algorithm -- Edge selection encoding -- Competition mechanism -- Learning interactive mechanism -- Learning parameter combination selection mechanism; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2021.107455
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: A new edge selection encoding method based on virtual component is presented. Novel competition and learning interactive mechanisms are proposed. A novel learning parameter combination selection mechanism is proposed. Comparative experiments and statistical analyses are conducted. Abstract: This paper presents an effective learning interactive genetic algorithm based on edge selection encoding (LIGA-ESE) for solving the assembly job shop scheduling problem (AJSSP), with the aim of minimizing the maximum completion time (makespan). In the LIGA-ESE, a novel ESE based on virtual components for producing every feasible solution generated with acceptable probability is proposed to improve the convergence efficiency. Based on the proposed competition and learning interactive mechanisms, two genetic algorithm populations from obverse and reverse schedules, respectively, compete to improve the utilization of computation resources and the quality of populations. In addition, the influences of parameters are considered, and a learning parameter combination selection mechanism based on historical optimization performance is proposed for further improving the convergence efficiency of the LIGA-ESE. Finally, a numerical simulation is conducted based on widely used AJSSP instances. Comparisons between the LIGA-ESE and existing algorithms and other LIGA-ESE versions demonstrate its effectiveness and superiority in solving the AJSSP.