2025

Conférence

2025 11th International Conference on Optimization and Applications (ICOA),1-6.

The Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) is designed to decompose multi-objective optimization problems (MOP) in scalar subproblems, collaboratively solved by sharing information among neighboring solutions. Among various approaches, The penalty bound intersection (PBI) method has been widely adopted because of its effectiveness in balancing convergence and diversity. Recent research has focused on adaptive parameter control during the search to improve the performance of MOEA/D. One of the main challenges in this context is resolving the exploration–exploitation trade-off in parameter selection. To address this, we propose a novel reinforcement learning-based parameter selection method, to solve Green Flexible Job-Shop Problem (Green FJSP), where the current configuration of algorithm parameters is modeled as a state, and each possible adjustment to these parameters is treated as an action. By using deep Q-networks (DQN) to train a policy that evaluates the Q-value of each given action in a state, the suggested method adaptively selects the most appropriate parameters in the optimization process, thereby improving the overall efficiency of the search and the quality of the solution.

@inproceedings{hamida2025self,
  title={A Self-Learning MOEA/D Framework Using Deep Q-Networks for Green Flexible Job-Shop Problem},
  author={Hamida, Maha Ben and Azzouz, Ameni and Said, Lamjed Ben},
  booktitle={2025 11th International Conference on Optimization and Applications (ICOA)},
  pages={1--6},
  year={2025},
  organization={IEEE}
}