Population-Level Hybridization between Roulette Wheel Selection and Tournament Selection for Particle Swarm Optimization

This paper introduces a population-level hybrid selection framework that combines the roulette wheel selection (RWS) and the tournament selection (TS) to pick promising exemplars to update particles, so as to ameliorate the capability of particle swarm optimization (PSO). As a result, a novel PSO, named PLHPSO is developed. PLHPSO distinguishes itself from the conventional PSO by employing a variety of exemplar selection strategies to steer the update of particles. To be precise, this framework first selects two different exemplars from all personally best positions and then it compares the particle to be updated with the two chosen exemplars. Next, different update strategies are designed to update the particle based on the comparison results. Subsequently, under this framework, totally six selection methods are developed, including TS, RWS, and their four hybridizations. With these selection strategies, particles in PLHPSO are able to learn from a wide range of directional exemplars, and hence the updating diversity of the swarm is hopefully increased. Experiments on the 50-D and 100-D CEC2014 benchmark problem suites demonstrate that PLHPSO with the six selection strategies outperform the traditional global and local PSOs. Particularly, among the six selection methods, the linearly decreasing hybridization between RWS and TS stands out as the most helpful one for PLHPSO to cope with optimization problems effectively.