Research On Feedback Multi-agent Genetic Algorithm Based On Multi-objective Optimization

In scientific research and engineering practice,many practical problems can usually be attributed to different types of mathematical programming problems,such as single-objective optimization problems and multi-objective optimization problems.With the development of science,technology and engineering industry,the abstracted mathematical problems are becoming more and more complex.Ordinary search algorithms can no longer meet the needs of complex problem processing.The bionic group algorithm represented by genetic algorithm is dealing with single and multiple targets.The superiority shown when optimizing problems makes it an important tool for solving target optimization problems.The advantages of genetic algorithm,such as high robustness,versatility and excellent global search ability,have attracted the attention of many scholars.Genetic algorithm is a general algorithm that does not require too much information when dealing with complex optimization problems,and has a significant effect on solving more complex combinatorial optimization problems.This paper studies single-objective optimization problems and multi-objective optimization problems and proposes several methods.The main innovations are as follows:(1)A feedback multi-agent genetic algorithm based on single-objective optimization is proposed.Aiming at the slow convergence speed of multi-agent genetic algorithm and the problem that the solution accuracy needs to be improved,a new feedback multi-agent genetic algorithm is proposed.The algorithm integrates the idea of uniform design,enriches the diversity of the initial population and verifies it;adding a feedback operator improves the convergence speed of the algorithm and greatly reduces the number of function evaluations.At the same time,the neighborhood competition,mutation and self-learning operators have been greatly improved,combined with arithmetic crossover,and binary competition to retain elite individuals.Highdimensional function optimization experiments show that the improved algorithm can avoid falling into the local extreme dilemma to a large extent,has a good global optimization ability and higher solution accuracy.(2)Propose a Pareto optimal solution set construction algorithm based on the queuing method.Aiming at the time and efficiency of constructing the Pareto optimal solution set for multiobjective decision-making problems,a Pareto optimal solution set construction algorithm based on the queuing method is proposed,and the nature of the non-dominant relationship and the related definition and construction of the Pareto optimal solution set are given.The theorem and its proof.The time complexity of the proposed algorithm is(rm N)and the worst time complexity is rm(2N-2m-1).The structure of the algorithm's construction set in the worst case is deduced.The comparison algorithm shows that when the proportion of non-dominated solutions is small(m/N=20%),the algorithm in this paper is better than the fast non-dominated sorting method in comparison times and CPU running time.When the proportion of non-dominated solutions is large(m/N =80%),the faster construction method of the algorithm in this paper and the arena match method are almost the same in the number of comparisons,but they have obvious advantages in CPU running time.(3)A feedback multi-agent genetic algorithm based on multi-objective optimization is proposed.The feedback multi-agent genetic algorithm based on single-objective optimization is integrated into the construction method used to construct the Pareto optimal frontier,and then the crowdedness distance method in the non-dominated sorting method is used for reference,and the concept of crowdedness is introduced to make the distribution of the constructed Pareto frontier solution More evenly.Finally,the experimental test is carried out through the commonly used 4double-objective test functions and 2 three-objective test functions.The test results show that compared with the comparison algorithm,there is a certain advantage in the uniform distribution of the solution.