An Improved Artificial Lemming Algorithm Integrating Non-Uniform Mutation and Q-Learning Adaptation for Underwater...

Area of Science:

• Artificial Intelligence
• Optimization Algorithms
• Metaheuristics

Background:

• Population diversity loss and premature convergence hinder Artificial Lemming Algorithm (ALA) performance in complex optimization.
• Existing ALA variants struggle to balance exploration and exploitation effectively.

Purpose of the Study:

• To develop an Improved Artificial Lemming Algorithm (IALA) that overcomes the limitations of the original ALA.
• To enhance optimization accuracy, convergence speed, and robustness through multi-strategy improvements.

Main Methods:

• Introduced a non-uniform mutation operator and nonlinear step-size strategy for improved local search.
• Incorporated a relative advantage learning strategy to enhance exploration capabilities.
• Integrated a Q-learning-based adaptive mechanism for intelligent switching among five behavioral modes.

Main Results:

• IALA achieved a Friedman mean rank of 1.25, outperforming original ALA and other state-of-the-art algorithms.
• Wilcoxon rank-sum test showed IALA significantly outperformed other algorithms in 118 out of 120 comparisons.
• Ablation studies confirmed the synergistic effect of all strategies, with Q-learning being the most critical.
• IALA demonstrated effective parameter tuning for underwater manipulator controllers.

Conclusions:

• The proposed IALA significantly improves upon the original ALA, offering enhanced optimization accuracy, convergence speed, and robustness.
• The Q-learning adaptive mechanism is crucial for balancing exploration and exploitation, leading to superior performance.
• IALA shows practical viability and efficiency in real-world engineering applications.