Hierarchical learning particle swarm optimization using fuzzy logic

Abstract

Particle swarm optimization (PSO) has been used extensively in numerical and engineering optimization problems in the last decades. However, due to drawbacks such as a single learning sample, PSO has the problems of loss of population diversity and easily falls into local optimum. To enhance optimization capability, a PSO based on fuzzy logic and hierarchical learning mechanism (FHPSO) is proposed. In FHPSO, the parameters are dynamically adjusted through fuzzy logic. The purpose of the fuzzy system is to generate appropriate parameters based on the performance metrics at each iteration, which better balances exploration and exploitation capability. Then particles are classified into different layers in terms of their fitness, and the particles in different layers perform different learning mechanisms. Each layer divides the particles into high-energy particles and low-energy particles. The high-energy particles in each layer are qualified to learn from the particles in the upper layer and the low-energy particles learn from the high-energy particles in their layer. This learning mechanism avoids all individuals to learn the global optimal individual at each iteration which will effectively reduce the speed and possibility of premature convergence and maintain population diversity. The FHPSO was compared with 8 well-known algorithms and 6 state-of-the-art PSO variants in the CEC 2022 and CEC 2021 test suites, respectively. The experimental results show significant performance of FHPSO. Simulation results for 5 complex engineering optimization problems and 3D path planning problem also show that the FHPSO can provides more competitive optimization results.