Gaining a Better Quality Depending on More Exploration in PSO

Abstract

We present a potential extension for particle swarm optimization (PSO) to gain better optimization quality on the basis of our agent-based approach of steering metaheuristics during runtime [1]. PSO as population-based metaheuristic is structured in epochs: in each step and for each particle, the point in the search space and the velocity of the particles are computed due to current local and global best and prior velocity. During this optimization process the PSO explores the search space only sporadically. If the swarm “finds” a local minimum the particles’ velocity slows down and the probability to “escape” from this point reduces significantly. In our approach we show how to speed up the swarm to unvisited areas in the search space and explore more regions without losing the best found point and the quality of the result. We introduce a new extension of the PSO for gaining a higher quality of the found solution, which can be steered and influenced by an agent.