The crowd framework for multiobjective particle swarm optimization

摘要

Multiobjective particle swarm optimization meets two difficulties—guiding the search towards the Pareto front and maintaining diversity of the obtained solutions—so a great number of improvements are possible. Our crowd framework systematically summarizes these improvements, extracts them into reusable strategies and categorizes them into modules by their optimization mechanisms. We introduce a number of new techniques within the modules. Strategies are compared first theoretically and then practically through amended ZDT series. We propose a sequence for module application based on the correlation between the modules. The resulting algorithms give incredible performance. Thus our crowd framework forms a new baseline for MOPSO.