Adaptive Multi-objective Particle Swarm optimization Based on Competitive Learning

In the evolutionary process of multi-objective optimization algorithm, the selection of the guide has a great influence on the flying state of particles. A more accurate guide selection strategy can enhance the convergence faculty and maintain the population's diversity. This article introduces an improved multi-objective optimization algorithm (AMPSO). Its core innovation is the introduction of competition. Firstly, the evolution process is divided into three states according to the different flight conditions of particles: convergence state, exploration state, and stagnation state. At the same time, the traditional g-best algorithm is divided into three kinds: c-gbst, d-gbest, and s-gbest. Then, according to the evolution state of particles, different leaders are selected. When the particles are in convergence state, c-gbest is selected as the leader to improve the convergence ability of the population. When the particles are in exploration state, d-gbest is selected as the leader to improve the diversity of particles. When the particles are in stagnation state, s-gbest is selected as the leader to guide the population out of the local area optimal. Compared with the other four modified multi-objective algorithms, we have carried out a series of experimental simulations of 10 benchmark functions. Experimental results reveal that the algorithm in this paper performs better.