Research And Application Of A Many-Objective Evolutionary Algorithm Based On Dynamic Scale Factors

No matter in scientific research or engineering application,optimization is one of the most common problems.At present,the research and application of optimization methods have penetrated into different fields,involving various aspects of production and scientific research,and has become a hot research hotspot by experts and scholars.The kind of optimization problem is called high-dimensional multi-objective optimization problems(MaOPs)when the optimization problem needs to be processed with not less than four objective functions.Moreover,because of its population-based solution method,a set of non-dominated solutions are obtained in a single run that are also parallel features.By combining GPU acceleration,the efficiency of the solution can be greatly improved,making it the most effective method for solving multi-objective optimization problems.Multi-objective evolutionary algorithm(MOEA)is a kind of global probability optimization heuristic search method formed by imitating biological evolution mechanism.It has been developed rapidly since the mid-1990 s.The multi-objective evolution algorithm starts with a set of initial populations and performs evolutionary operations on the population,such as selection,crossover,and mutation.After many generations of evolution,better individuals were continuously obtained and gradually approached the Pareto boundary of multi-objective optimization problems.The multi-objective evolutionary algorithm can process a group of potential solutions in parallel and is insensitive to the shape and continuity of the Pareto front of the problem.Problems with more than three objective functions are called many-objective optimization problems(MaOPs).At present,many traditional multi-objective evolutionary algorithms have obvious defects in dealing with many-objective problems.In order to solve this problem,this paper proposes a many-objective evolutionary algorithm based on dynamic scale factor,and combines fitness landscape analysis with many-objective evolutionary algorithm.Through experimental comparison,the characteristics are solved and its effectiveness is verified.The main achievements and innovations of this paper are as follows:Firstly,this paper proposes a new distance factor.Clustering methods are used to classify populations,and then the approximate Pareto frontiers of several segments are fitted by regression.Then the Euclidean distance of the approximate Pareto frontier of the class is calculated.Through this distance,the individual dispersion can be determined to provides better diversity selection conditions for evolutionary optimization.Secondly,for the congestion density distance and the new distance factor,the dynamic scale factor is introduced,and a many-objective evolutionary algorithm based on dynamic scale factor is proposed.The dynamic scale factor can dynamically adjust the influence ratio of the new distance factor and the crowded density distance during the evolution so as to balance the diversity and convergence of the evolutionary process.By solving the classical multi-objective problem and comparing with some other classical multi-objective algorithms,the rationality of the new algorithm is verified.Finally,the multi-objective evolutionary algorithm is used to represent the landscape features of fitness landscapes.By using adaptive landscape analysis strategies such as adaptive distance correlation and random walk sampling,the landscape characteristics of many-objective evolutionary algorithms are analyzed.