An improved genetic algorithm based on reinforcement learning for aircraft assembly scheduling problem

• Published in: Computers & industrial engineering Vol. 193; p. 110263
• Main Authors: Wen, Xiaoyu, Zhang, Xinyu, Xing, Hongwen, Ye, Guoyong, Li, Hao, Zhang, Yuyan, Wang, Haoqi
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2024
• Subjects: Aircraft assembly; Genetic algorithm; Markov decision making; Reinforcement learning; Aircraft assembly; Markov decision making; Genetic algorithm; Reinforcement learning
• ISSN: 0360-8352
• DOI: 10.1016/j.cie.2024.110263

•A mathematical formulation for the aircraft assembly scheduling problem based on multiple constraints.•An improved genetic algorithm based on Q-learning to match the multiple constraints problem.•Developed a Markov Decision Process (MDP) model to regulate crossover and mutation probabilities in the algorithm. The assembly work of aircraft occupies a significant portion of the time in aircraft manufacturing. The aircraft assembly operations are highly challenging in the generation of production plans due to multiple constraints in the scheduling process. In addressing the aircraft assembly scheduling problem with the objective of minimizing the duration of assembly operations, a mathematical model has been established. This model incorporates characteristics such as job sequencing constraints, resource constraints, spatial constraints, and others, the problem-solving process is divided into two stages: determining the priority sequence of tasks and establishing the start times for each task. Proposing an improved genetic algorithm based on Q-learning, referred to as QIGA (Q-learning-based Improved Genetic Algorithm), utilizing Q-learning to dynamically adjust the crossover and mutation probability parameters during each iteration to enhance the algorithm’s convergence speed and search capabilities. Constructing a Markov Decision Process (MDP) model to dynamically adjust the crossover and mutation probability in the problem, proposing a state set partition rule based on the population’s fitness values, setting the action set to represent crossover and mutation probabilities in different intervals. Additionally, designing reward rules based on the population’s performance. Finally, the effectiveness of the proposed algorithm is verified with a real case containing 44 assembly operations, and the results show that the algorithm can solve the aircraft assembly scheduling problem well. Validating the feasibility and optimization effectiveness of the algorithm through various aircraft assembly scheduling cases of different scales. The experimental results indicate that, compared to traditional meta-heuristic algorithms, the proposed algorithm demonstrates superior optimization performance in aircraft assembly scheduling problems. It effectively reduces the total duration of aircraft assembly operations.