Knowledge Reconstruction for Dynamic Multi-objective Particle Swarm Optimization Using Fuzzy Neural Network

Many real−world applications are dynamic multi−objective optimization problems (DMOPs). The transfer of knowledge in the evolutionary process is believed to have advantages in solving DMOPs. However, most existing works can hardly be focused on the effectiveness of knowledge, which may lead to the negative transfer to degrade searching performance of the population. To address this issue, a knowledge reconstruction (KR) method is proposed for dynamic multi−objective particle swarm optimization (DMOPSO) using fuzzy neural network (FNN). The contributions of the proposed KR−DMOPSO are threefold: First, a knowledge extraction method, using a FNN model, is developed to obtain the domain knowledge of two successive Pareto optimal sets when dynamic occurs. Then, the domain knowledge can be applied to explore the evolutionary tendency. Second, a knowledge evaluation mechanism, based on the diversity and convergence of non−dominated solutions, is devised to select the domain knowledge. Then, the effective knowledge can be achieved. Third, a knowledge reconstruction strategy is designed to obtain the suitable domain knowledge. Then, this knowledge can be used to adapt to dynamic environments to improve the searching performance of the population. Finally, the proposed KR−DMOPSO is compared with other advanced dynamic multi−objective optimization algorithms (DMOAs). The results show that the proposed KR−DMOPSO is superior to other compared algorithms.