Research On Improvement And Application Of Multi-objective Optimization Algorithm Based On Decomposition

A large number of engineering application problems,problems in scientific research,and problems in practical applications can be converted into multi-objective optimization problems.Research on algorithms for solving multi-objective optimization problems has always been a hot topic.In the past two decades,with the in-depth study of evolutionary algorithms such as genetic algorithms and multi-objective optimization problems,evolutionary algorithms have become the mainstream to solve the multi-objectives optimization problems.The multi-objective evolutionary algorithm is a kind of heuristic algorithm,which has the advantages of versatility,parallelism and multiple problem solutions in one run.Multi-objective Evolutionary Algorithm based on Decomposition(MOEA/D)is a recently proposed multi-objective evolutionary algorithm that decomposes multi-objective optimization problems into multiple simple single-objective optimization problems through decomposition functions.The problem,which reduces the complexity of solving multiobjective optimization problems and high-dimensional multi-objective optimization problems(Many-objective).However,there are still great challenges in solving difficult multiobjective optimization problems and application levels.Therefore,effective improvement of the MOEA/D framework and its application to practical problems will have very important theoretical significance and application value.This paper is mainly based on the improvement of MOEA/D framework and its application research.The main work are summarized as follows:1)In the multi-objective optimization algorithm,the two points of great concern are the convergence and diversity of the Pareto solution set.In the optimization process,the selected Pareto solution set must be as close as possible to the Pareto front.It should be distributed as evenly as possible along the Pareto front.In the MOEA/D algorithm framework,if the diversity of the Pareto solution set is not easy enough to cause multiple sub-problems to correspond to the same solution,this is actually not conducive to the overall optimization of the algorithm,and finally the Pareto solution set cannot be maintained well.The diversity is also difficult to achieve better convergence.Aiming at this problem,this paper first discusses the correspondence between the sub-question set and the Pareto solution set.Then,based on the vector corresponding to the sub-question,the concept of regional division of the target region of the solution is described.According to the concept,an adaptive region adjustment strategy is proposed,which is used to balance the convergence and diversity of the Pareto solution set.Then the strategy is embedded in the MOEA/D algorithm framework to propose a simple but effective algorithm implementation.Finally,in order to prove the effectiveness of the proposed algorithm,a comprehensive comparison experiment and parameter experiment are designed.The experimental results show the effectiveness of the proposed algorithm.2)In the framework of MOEA/D algorithm,the optimization process and difficulty between sub-problems are not the same,so the computing resources required by different sub-problems and the order of computing resources between sub-problems are also different.However,all sub-problems are assigned the same resources and are treated differently in the MOEA/D algorithm framework.In fact,the rational allocation of computing resources between sub-problems helps to reduce the wastage of computing resources and generally improves the performance of the algorithm.Aiming at this problem,this paper proposes a resource allocation strategy based on the relationship between MOEA/D subproblems.The strategy maintains a probability vector by updating the replacement relationship between the sub-questions,and the vector is used to guide which sub-problems are selected for the next iteration optimization.Based on the exploration of the role of boundary sub-problems in the whole optimization process,a priority boundary sub-problem strategy is proposed to further improve the resource allocation strategy.Then the strategy is embedded in the MOEA/D algorithm framework to propose a stable algorithm.Finally,the comprehensive comparative experiment and parametric experimental research show that the algorithm has better performance.3)Rapid mutations in influenza viruses have enabled them to escape group immunity,which has become a key challenge in influenza vaccine design.It is important to predict the evolution of influenza antigens and to identify new antigenic variants in a timely manner.However,traditional experimental methods such as hemagglutination inhibition(HI)determination of vaccine strains are time and labor intensive.In this paper,a new multi-objective optimization based method for the determination of antigenicity of influenza virus based on hemagglutinin(HA)sequence is proposed.This method combines linear regression model,polynomial regression model and antigen mapping to utilize influenza.The HA1 sequence similarity quantifies the antigen distance and is optimized by the multi-objective optimization algorithm proposed in this paper.Firstly,a new low rank matrix completion model was proposed.Based on the partially exposed antigen distance,based on the similarity of the HA protein sequence and the similarity of the vaccine based vaccine,the antigen distance between the antigen and the antiserum was inferred.Restore missing values or error values in the HI table.Then,based on this,the antigen distance between viruses was quantified as a response to the regression model,and a regression prediction model was constructed.Finally,through a large number of comprehensive computational experiments,the advantages of the proposed algorithm are presented.