Global Optimization using Random Adaptive Backtracking Particle Swarm Optimization (RAB-PSO)

Abstract

Optimization algorithms, like the Particle Swarm Optimization (PSO), often suffer from premature convergence, providing poor convergence quality and slow convergence rates. In addition, striking a balance between exploration and exploitation adds complexity to its implementation. Moreover, while the algorithm's simplicity with a few parameters is advantageous for ease of use, it poses a significant challenge for improvement. This work presents a modified PSO variant, the Random Adaptive Backtracking Particle Swarm Optimization (RAB-PSO) algorithm. This algorithm combines three complementary modifications to address the limitations of PSO. Its main objective is to improve convergence quality while minimizing iteration counts required for achieving global minima. The first modification applied a delta parameter inspired by golf ball movements and incorporated it into the velocity formula. The second modification ensures that the inertia weight (ω) is adjusted to an S-shaped descending function to balance the search process. The third modification introduces a novel update strategy, enabling particles to backtrack when trapped, improving the algorithm's capacity to escape local optima and address premature convergence. These enhancements yield significant improvements, with the RAB-PSO algorithm demonstrating superior convergence performance on four test functions, requiring fewer iterations to reach optimal solutions.