Research On Large-scale Multi-objective Optimization Algorithm Based On Evolutionary Computation

Multi-objective optimization problems(MOPs)exist widely in scientific research and engineering.Researchers have proposed many classical multi-objective optimization evolutionary algorithms(MOEAs)to solve MOPs,and have achieved certain results.However,MOEAs cannot solving large-scale multi-objective optimization problems(LSMOPs)with more than 100 decision variables.The reasons are as follows: Firstly,the search space of the algorithm will increase exponentially when the decision variables of the optimization problem increase.But the vast majority of MOEAs optimize the decision variables as a whole,which will lead to a sharp deterioration in the performance of MOEAs.Secondly,the larger the number of decision variables,the more complex the interaction between the variables.In the objective space,a large number of decision variables will have a certain impact on the convergence and distribution of the solution.Therefore,it is necessary to design a large-scale multi-objective evolutionary algorithm to solve LSMOPs.Based on this,this paper carries out the following work:(1)A large-scale multi-objective optimization algorithm using neighborhood adaptive strategy of differential evolution(NAS-MOEA)is proposed.All variables are divided into diversity variables and convergence variables according to analyzing the domination relation among the disturbed solutions of the decision variables,which makes the variable classification more accurate and improves the efficiency of the evolutionary process.Then,principal component analysis denoising for convergence variables.Not only the main information of the original data is preserved,but also the subcomponents are not related,thereby reducing the computational cost and obtaining better results.Besides,the alternately evolution strategy is used to optimize the convergence variables and the diversity variables,which balances the relationship between the convergence and diversity of the evolutionary population and avoids the algorithm from local optimum.The experimental results demonstrate that the proposed algorithm has significant advantages in convergence speed and optimization accuracy.(2)A large-scale multi-objective optimization algorithm based on mixed decision variable analysis(MDA-MOEA)is proposed.MDA-MOEA has designed a new mixed decision variable analysis method in order to improve the classification accuracy of decision variables and reduce the problem of convergence difficulties caused by misclassification.Instead of setting the difference point between the convergence variable and the diversity variable,the control attribute of the decision variable is quantified as a numerical representation using the control property sorting formula.In addition,in order to improve the efficiency of the algorithm to solve LSMOPs,MDA-MOEA also added a penalty strategy to correct the deviation of the fitness function,thereby shortening the runtime of the algorithm.The experimental results show that the algorithm proposed in this paper exhibits good performance,especially on the test problem with more than 300 decision variables,and has certain performance advantages in balancing convergence and diversity.In summary,this paper proposes two different large-scale multi-objective optimization algorithms based on evolutionary computing for LSMOPs,the problem of inaccurate classification in the decision variable analysis is deeply studied,a more targeted optimization method is designed in the optimization stage.The performance of the proposed algorithm is verified in the relevant test functions.