Hyperparameter Control Using Fuzzy Logic: Evolving Policies for Adaptive Fuzzy Particle Swarm Optimization Algorithm

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Heuristic optimization methods like particle swarm optimization (PSO) require careful parameter tuning for optimal performance.
• Adapting algorithm parameters during the optimization process is a key challenge in improving heuristic methods.

Purpose of the Study:

• To develop a novel approach for designing parameter adaptation strategies in heuristic optimization using continuous fuzzy feedback control.
• To systematically create and evaluate a diverse set of fuzzy-controlled PSO algorithms.

Main Methods:

• Implemented a continuous fuzzy feedback control framework to dynamically adjust PSO parameters.
• Optimized fuzzy processes beforehand using a training benchmark to maximize performance.
• Generated 127 distinct fuzzy PSO algorithms with up to seven fuzzy-controlled parameters.

Main Results:

• The newly developed fuzzy PSO algorithms demonstrated superior performance over traditional PSO and existing parameter control variants.
• Performance was validated in the Congress on Evolutionary Computation (CEC) 2020 competition for single-objective bound-constrained numerical optimization.
• Two specific fuzzy controls showed strong efficacy and dependability in real-world scenarios from CEC 2011.

Conclusions:

• Continuous fuzzy feedback control offers an effective mechanism for adaptive parameter tuning in PSO.
• The proposed fuzzy PSO algorithms represent a significant advancement in optimization performance, outperforming established methods.
• The framework is robust and applicable to both benchmark numerical optimization and real-world problems.