An Improved Multi-objective Particle Swarm Optimization with Mutual Information Feedback Model and Its Application

Multi-objective optimization problems have been widely studied in the evolutionary computation domain, and many algorithms with evolutionary computation have been successfully adopted to solve multi-objective optimization problems. The information feedback model is an effective method to generate new individuals by combining previous and current information. It has been successfully used to improve the performance of particle swarm optimization to solve single-objective optimization problems. However, the current information feedback model has at least two drawbacks. First, the function evaluation number is increased for all individuals should be evaluated twice in a generation. Second, the model cannot directly be used to solve the multi-objective optimization problems because the update coefficient of the feedback model is difficult to determine for many objectives. To solve these problems, an information feedback model based on mutual information is developed to directly solve the multi-objective optimization problems. The model can be easily combined with any optimization algorithm. The model was used in particle swarm optimization to form a new multi-objective particle swarm optimization in this study. In this method, the external archive of multi-objective particle swarm optimization was built by non-dominated sorting method, the basic operator of multi-objective particle swarm optimization and the information feedback model based on mutual information were used to update individuals with different characters. The convergence and diversity of the population were better balanced. The effectiveness of the method was tested on non-constrained, constrained, and real-world problems. The results of the method were compared to those of some state-of-the-art multi-objective optimization problems. The comparisons indicate that the proposed algorithm is challenging in terms of the generational distance and spacing of solutions in multi-objective optimization problems.

Automated summary

The study presents an information feedback model based on mutual information to directly handle multi-objective optimization problems, which can be easily integrated with any optimization algorithm, improving the balance between convergence and diversity in the population.