A fast-flying particle swarm optimization for resolving constrained optimization and feature selection problems

Particle Swarm Optimization (PSO) is popular because of its ease of use and few parameter tunings. Also, updating particle velocity using the global and particles’ personal best positions as guides is appealing. However, such an update process can cause “Oscillation” and “Two-step forward, One-step backward” phenomena, negatively affecting PSO’s performance during optimization. Concurrently, it lacks adequate exploitation and exploration–exploitation balance skills, causing premature convergence. Hence, we propose a new hybrid PSO named Fast-flying PSO (FF-PSO) to address PSO’s issues. FF-PSO engages only one guide produced by a quantization technique to update velocity, significantly lessening PSO phenomena’s adverse impact and enhancing exploitation. Subsequently, it uses a dynamic adaptation of the search dimension strategy to update particle positions. Consequently, refined convergence and solution quality are achieved with better local optima avoidance ability. FF-PSO’s efficacy was assessed over 16 basic and Congress of Evolutionary Computation (CEC)-2017 competition’s benchmark functions relating to some of the latest algorithms. Next, the practicability of FF-PSO is verified by tackling CEC-2011’s constrained optimization and feature selection issues. An inspection of the results of distinct algorithms, including statistical analysis, shows that FF-PSO is securing the top spot among the contenders in over 60% of the problems from each category with some percentage of ties. Thus, FF-PSO may be a more fruitful technique in constrained and possibly other global optimization and feature selection tasks with intensified exploitation and refined convergence.