Novel multiobjective particle swarm optimization based on ranking and cyclic distance strategy

To effectively improve the convergence and diversity of the multiobjective particle swarm optimization (MOPSO), we proposed a novel MOPSO based on ranking and cyclic distance (RCDMOPSO) that comprehensively considers the spatial target and congestion information of particles. RCDMOPSO introduced a method namely global proportional ranking (GPR) which differs from nondominated ranking under the Pareto framework, and designed a novel external archive maintenance and the global selection strategies of learning sample by combining GPR with cyclic distance. In this paper, RCDMOPSO together with eight classic and state-of-the-art algorithms were examined on ZDT, UF, and DTLZ series to test functions. The results show that RCDMOPSO is highly competitive in achieving the objectives of both convergence and diversity. RCDMOPSO outperformed other popular algorithms such as MOPSOs and multiobjective genetic algorithms based on comprehensive performance evaluation indicators inverted generational distance and hypervolume, thus supporting that RCDMOPSO is an effective approach to tackle multiobjective optimization problems.

Automated summary

In this paper, a novel multiobjective particle swarm optimization algorithm (RCDMOPSO) is proposed, which comprehensively considers spatial target and congestion information of particles. RCDMOPSO introduces a method called global proportional ranking (GPR) and combines it with cyclic distance to design novel external archive maintenance and global selection strategies. Experimental results show that RCDMOPSO outperforms other popular algorithms and is effective in tackling multiobjective optimization problems.