Hybrid Differential Evolution Algorithm Integrating Operations Research and Reinforcement Learning approaches
Panagiotis G. Giannopoulos, Gregory P. Chondrokoukis, Thomas K. Dasaklis · 2024
The integration of Operations Research (OR) and Reinforcement Learning (RL) techniques is becoming increasingly important in improving the applicability of Artificial Intelligence (AI) models.In this paper, we introduce a novel hybrid Differential Evolution Algorithm (DEA) that utilizes Q-Learning (QL) to dynamically adjust crossover and mutation probabilities.In this approach, the selection operator precedes the mutation and crossover operators and is enhanced by the Preference Ranking Organization Method for Enrichment of Evaluations (PROMETHEE II), which considers multiple criteria such as fitness, diversity, and stability.Simulated Annealing (SA) is employed to further optimize the elitism mechanism by adjusting thresholds for retaining elite solutions.When evaluated on complex multi-dimensional objective functions previously used in OR-related problems, the hybrid method significantly improves performance and convergence, outperforming both standard and QL-enhanced DEAs.Sensitivity analysis confirms the model's robustness, highlighting its potential for solving multi dimensional problems across various domains.