Using Crowding Distance to Improve Multi-Objective PSO with Local Search
Ching-Shih Tsou, Shih‐Chia Chang, Po-Wu Lai · 2007
It is well known that local search, even in its simplest form, prevents search algorithms from premature convergence and, therefore, possibly drives the solution closer to true Paretooptimal front. A local search procedure and a flight mechanism both based on crowding distance are incorporated into the MOPSO, so called MOPSO-CDLS, in this paper. Computational results against ZDT1-4 problems show that it did improve the MOPSO with random line search in all aspects except the execution time. Local search in less crowded area of the front not only reserves the exploitation capability, but also helps to achieve a well-distributed non-dominated set. Global guides randomly selected from the less crowded area help the particles dominated by the solutions in this area to explore more diverse solutions and in a hope to better approximate the true front. This study intends to highlight a direction of combining more intelligent local search algorithms into a Pareto optimization scheme. Mechanisms based on crowding distance employed here did not explicitly maintain the diversity of non-dominated solutions which is its original intention, but they indeed facilitate the possibilities of flying towards the Paretooptimal front and generating a well-distributed non-dominated set. Further researches include comparisons with other multi-objective evolutionary algorithms and accommodating constraints-handling mechanism in the Pareto optimizer.