Research And Application Of Multi-Objective Differential Evolution Mutation Strategy

With the widespread application of artificial intelligence technology,the swarm intelligence algorithm has been in-depth research and rapid development.Among the numerous swarm intelligence algorithms,relying on its unique and rich mutation strategies,the differential evolution algorithm has been successfully applied to various optimization problems,so it has gradually become a popular research algorithm.However,the current mutation strategy research is limited to single-objective optimization and simple multi-objective optimization,and its application on complex multi-objective optimization problems need to be studied in depth.This thesis takes the multi-objective differential evolution mutation strategy as the research object,and faces complex multi-modal multi-objective optimization problems and constrained multi-objective optimization problems,and completes the corresponding mutation strategy and multi-objective optimization algorithm design,algorithm performance verification,and practical application.The main works of this thesis are as follows:1.The importance of the differential evolution algorithm and the related mutation strategies is presented by introducing the research background,and the main studies of this thesis are introduced by analyzing the defect of the current multi-modal multi-objective optimization algorithms and constrained multi-objective optimization algorithms.Furthermore,by reviewing the existing single-objective mutation strategies,multi-objective mutation strategies,and the related design methods,the research ideas for designing multi-objective mutation strategies for the problems to be solved in this thesis are provided.2.A cluster-based multi-modal multi-objective differential evolution algorithm is proposed to address multi-modal multi-objective optimization problems.In this algorithm,a special crowded distance method based on clustering technology is first designed to measure the overall crowded degree of individuals in the decision space and the objective space.Then,one distance-based exemplar selection mechanism is designed.Combining these two methods,the DE/current-to-exemplar/1 mutation strategy is proposed,which guides individuals to evolve towards the direction of high-quality and sparse solutions,so as to maintain diversity and generate high-quality offspring.The experimental results on the multi-modal multi-objective benchmark test set show that the proposed algorithm can effectively improve the population diversity and preserve the equivalent multi-modal solutions.Furthermore,for the location selection optimization,the designed algorithm can find all multi-modal Pareto optimal solutions and provide decision makers with high-quality alternative feasible schemes.3.A constrained multi-objective optimization algorithm using the relationship between unconstrained and constrained Pareto fronts is proposed to tackle constrained multi-objective optimization problems.In this method,the entire evolutionary process is divided into the learning stage and the evolution stage.In the learning stage,effective learning strategies and classification method are designed to explore the relationship between the two fronts.According to the obtained relationship,different evolution strategies are designed in the evolution stage to make better use of objective information.For different types of problems,two mutation strategies,DE/transfer/1and DE/current-to-opbest/1,are proposed in this algorithm.DE/current-to-opbest/1 is an improved version of DE/current-to-exemplar/1.The experimental results on the benchmark test sets prove the effectiveness of the proposed strategies and the superiority of the algorithm.Furthermore,the proposed algorithm is applied to solve the practical constrained multi-objective optimization problem,i.e.,vibration table design problem.Compared with other algorithms,the proposed algorithm finds more feasible Pareto optimal solutions and effectively mines the potential information in the decision space of the problem.