基于在线归档技术的多目标粒子群算法

Online Elite Archiving in Multi-Objective Particle Swarm Optimization

提出一种基于在线归档技术的新型多目标粒子群优化算法.使用外部集归档,在归档粒子中采用适应值共享技术选出全局最优位置,使得种群多样性得以维持;在粒子群的进化过程中,使用在线归档策略,将归档的粒子合理地引入下一代的种群,淘汰原种群中的不良粒子,从而保证进化过程中种群的优良性.用Zitzler的两个多目标测试函数评价算法的性能.结果表明,该算法能快速收敛到Pareto非劣最优目标域,并且解集沿着Pareto非劣最优目标域有很好的扩展性.

A multi-objective particle swarm optimization algorithm based on online elite archiving is proposed. The elite particles are put into repository. Fitness sharing is adopted to select global best position from the repository, thus the diversity of the population is guaranteed. In the course of evolution the online archiving technique is adopted, namely the elite particles in the repository are introduced into the population and inferior individuals are eliminated. Thus an excellent population is ensured. Two Zitzler functions are used to evaluate the performance of the proposed approach. Experiments demonstrated that the proposed method can rapidly converge and can effectively generate a satisfactory approximation of the Pareto front.