An adaptive decision sand cat swarm optimization algorithm combined with Q-learning for solving engineering problems

Abstract Problem Despite the advantages of multi-strategy improvements in enhancing the performance of metaheuristic algorithms, traditional strategy selection approaches often lack the sophistication necessary to maintain an effective balance between exploration and exploitation. Aim This study aims to improve the Sand Cat Swarm Optimization (SCSO) algorithm, which is recognized for its scalability and optimization potential. By incorporating multi-strategy enhancements, the goal is to elevate its performance on complex problems and facilitate dynamic strategy adjustments across different operational phases. Methods Thereby, four improved position update strategies were developed: adaptive weight factors, population position information analysis, cosine oscillation, and multi-scale perturbation, to enhance SCSO's search capability and prevent local optima. Additionally, the Q-learning strategy selection method was improved, and an adaptive decision-making framework was developed to choose the optimal update strategy for each sand cat, effectively balancing exploration and exploitation while preserving population diversity. Results Through validation with the CEC2022 benchmark functions and five engineering optimization problems, demonstrated that the improved algorithm significantly enhances solving accuracy and stability on complex problems compared to six other advanced swarm intelligence algorithms and non-swarm algorithms. Conclusion Despite a slight impact on convergence speed, the original simplicity of the SCSO algorithm is preserved, facilitating future expansions and maintenance.