Font Size: a A A

Research On Evolutionary Multi-objective Algorithm Based On Non-dominated Sorting

Posted on:2021-04-15Degree:MasterType:Thesis
Country:ChinaCandidate:B N YuFull Text:PDF
GTID:2428330647962026Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Most of the problems we encounter in daily life today are composed of several factors that affect or even exclude each other.This type of problem that requires simultaneous processing of multiple target parameters and reaching a balanced optimal state is called a multi-objective optimization problem.With the rapid development of contemporary intelligent production technology and automatic control technology,this type of problem generally exists in industry,medical treatment,personalized recommendation,etc.,and how to efficiently and accurately obtain the optimal solution set of this type of problem has become industrial application and science.The key subject to be solved urgently in the research field.Since the introduction of genetic algorithm in 1975 to the derivation of various algorithms today,evolutionary algorithms have been widely used to solve multi-objective optimization problems.The characteristics of the root play multi-objective evolutionary algorithm's core ideas can be roughly divided into the following three types: One is the multiobjective evolutionary algorithm based on the non-dominated relationship,that is,the Pareto dominant relationship between individuals is used to compare and select the optimal solution set.The second is based on the evolution of decomposition ideas,usually using traditional decomposition methods to decompose the required problem,and then using appropriate optimization methods to obtain the solution of the sub-problem to generate a solution set.The last category is the hybrid algorithm,which combines different advantages of different algorithms to solve compound optimization problems.Based on the scientific research of existing evolutionary algorithms,this paper analyzes the current theoretical thinking of solving problems,and conducts research on hybrid algorithms to carry out the following main work.First,the individual cumulative control strategy is used in the non-dominated genetic algorithm to propose a new algorithm IGNSGA,that is,the individual's Pareto ranking value in the current generation and the number of surrounding dominant individuals are summed,and the ranking sum value is recorded as the cumulative ranking value.The larger the individual cumulative ranking value,the higher the individual's excellence.Compared with Pareto control,this strategy can increase search pressure moderately,help to filter out the poor individuals in the same level when Pareto is ranked,save excellent Pareto solutions,and improve the convergence and distribution of IGNSGA algorithm.Second,for the critical level layer in which the elite solution and the candidate solution to be determined coexist,an adaptive grid division mechanism is introduced: the individual distribution of the population is divided into grids,and the mixed distance between the individuals in the grid is compared.The smaller the distance,the better.By replacing the iterative process with this,the effects of reducing calculation,saving time and fast search speed can be achieved relatively.Third,considering that there may be points where the mixing distances are equal in the same grid,a weight preference vector is proposed for this situation: that is,setting weight preference according to actual requirements in multiple solutions of equal excellent degree,calculating the preference mixing distance,distance The larger the solution is selected as the optimal solution,so as to better integrate the actual application and achieve the expected effect of the optimal solution being more humanized.Finally,propose a hybrid algorithm WPA-NSGA,that is,select the head wolf,detective wolf,and fierce wolf according to the Pareto dominance relationship in the non-dominated genetic algorithm,and then use the wolf group operator to screen the excellent solution and store it in the external archive.Using the good convergence and stability of WPA itself,to a certain extent,it improves the Pareto dominance relationship.When facing high-dimensional MOP,the Pareto hierarchical sorting is too time-consuming due to the weak search pressure,so that the algorithm can solve high-dimensional and complex optimization problems.
Keywords/Search Tags:Multi-objective evolutionary algorithm, Pareto domination, cumulative sorting strategy, adaptive meshing, WPA
PDF Full Text Request
Related items