Research And Application Of Multi-Objective Particle Swarm Optimization Algorithm Based On Decomposition

Particle Swarm Optimization (PSO), a recently proposed evolutionary algorithm, is very popular among engineering practice and multi-objective optimization research, because of its superior performance and simple principles. However, the increasing number of variable dimensions and the irregular shape of Pareto frontier have caused seriously performance degradation in solving multi-objective optimization problems with current PSO. To solve the two problems, this paper introduce decomposition method in researching multi-objective particle swarm optimization algorithm and apply the cooperative co-evolution and the bipolar preferences to decompose the variable space and the Pareto front. And the main contributions of this paper are as follows:1. For solving the problem of easily trapped into local optimal, the mutation operator, the inertia weight and the elite population are first investigated through theoretical analysis. Specifically, a non-uniform mutation strategy is introduced to promote particles escape from local optimal. Meanwhile, the inertia weight is set to be linearly decreasing to balance the algorithm’s capability of global search and local search. Finally, the classical elite population management strategy will be applied to manage the non-dominated solutions.2. To solve the problem of the performance degradation of algorithm along with the increasing decision variables, this paper are going to study optimization methods based on decomposition and find the mathematical characteristics of variable interaction of multi-objective function. Then this paper will come up with an optimized multi-objective PSO via combining cooperative co-evolution and developing methods to find the interactive variables of multi-objective function with less computation in the searching procedure. Finally, the proposed algorithm will be compared with several standard test functions. The results of the experiments show that our new approach has better performances compare with MOPSO and NSGA-II in terms of diversity and convergence.3. For the concept of bipolar preferences, we consider it as a decomposition of the Pareto frontier which is a totally different interpretation. In the application research, according to decision makers’preferences (positive preferences:high detection rate and low false positive rate; negative preferences:low detection rate and high false positive rate), this paper introduces the bipolar control strategy and uses sliding step and the scale step as the optimizing parameters of Bipolar Preference Multi-objective Particle Swarm Optimization (BPMOPSO). The experiment results show that the performance of the new method is greatly improved, i.e., the false positive rate decreases sharply while the detection rate improves significantly, and the efficiency of the algorithm increases as well.