Abstract:
Aiming at the leader particle selection problem, we present a multi-objective particle swarm optimization algorithm based on density clustering, termed as DN-MOPSO. Firstly, the particles composed of evolutionary population and external archive are divided into several classes by using density clustering technique in decision variable space. And then, the particles in the evolutionary population select their respective global leader particles according to the classification results so as to achieve the balance between global search capabilities and local search capabilities. Moreover, when the non-dominated solutions exceed the maximum limit, the improved circular crowed distance will be utilized to sort the external archive so that the obtained solution set can be well the distributed. In addition, the Gauss mutation operator mechanism is introduced into the iterative process to ensure the proposed algorithm jumping out of premature and promote the search capacities. Finally, the proposed DN-MOPSO algorithm is tested on several WFG benchmarks functions and compared with some classic multi-objective optimization algorithms. Simulation results show that DN-MOPSO algorithm can achieve better approximation to the optimal Pareto front and get well-spread Pareto solutions. Moreover, this strategy in this paper can not only enlarge the search scope of the DN-MOPSO algorithm, but also improve the search accuracy, meanwhile, maintain the diversity of non-dominated solution. Therefore, the convergence and diversity of the solutions obtained by DN-MOPSO algorithm are significantly improved. These simulation results also indicate that DN-MOPSO algorithm is highly feasible and competitive for solving multi-objective optimization problems.