Examination of benefits of personal fitness improvement dependent inertia for Particle Swarm Optimization

Abstract

Since its invention, Particle Swarm Optimization (PSO) has received significant attention in the optimization community, which spawned numerous PSO modifications, variations and applications. However, most of the PSO improvements come with impaired simplicity and increased computational cost of the method. As an effort to advance the PSO performance through enhanced particle awareness of its own fitness, a novel PSO modification based on personal fitness improvement dependent inertia (PFIDI) is proposed. The PFIDI technique used in the paper employs a straightforward and elegant switch-like condition on inertia which turns off a particle’s inertia when the particle stops advancing in a direction of better fitness. Considering the effects of this technique on the particle movement logic, the method is called “Languid PSO” (LPSO). So as to attain a reliable assessment of the effects of PFIDI as implemented in LPSO, a massive computing effort was exerted for the benchmark testing, in which LPSO accuracy was compared to standard PSO accuracy on 30 test functions (CEC 2014 test suite), three problem space dimensionalities (10, 20 and 50), and a wide range of PSO parameters. The results clearly show the advantages of PFIDI-enabled LPSO, which predominantly outperforms standard PSO, both across all parameter combinations and for best-achieving PSO parameters. The success of the proposed PSO modification, coupled with its elegance and computational simplicity (less than 1.1 % increase in computational cost over standard PSO), indicates that fitness-based inertia may represent a rewarding approach in the PSO research.