Research And Application Of Multi-objective Differential Evolution Algorithm Based On Pareto

Evolutionary computation is a random search algorithm which imitates natural screening and natural evolution of biology.Because it is suitable for solving highly complex uncertain questions,it has obtained quite a broad application field,and it has good universality.After unremitting efforts,many scholars around the world have created a variety of excellent optimization algorithms,and differential evolution algorithm is one of the best.In this thesis,based on the principle of pareto environment selection strategy,the single objective,multi-objective and constrained differential evolution algorithm are improved respectively.The main research contents are as follows :(1)Analyze the basic theory of differential evolution algorithm,briefly describe the evolution process of differential evolution algorithm and the operator in the algorithm.(2)As a precursor,the single objective differential evolution algorithm is improved in three aspects.First,the initialization uses the reverse learning method to find the parent population with better fitness value;the second is the adaptive improvement of mutation and crossover operator;thirdly,it is integrated with the improved genetic evolutionary algorithm of niche technology using chaotic initialization and clustering exclusion mechanism.Comparing the new algorithm with the other four algorithms through the test function,the results show that the performance of the new algorithm is better than the other four algorithms.(3)The basic knowledge of multi-objective optimization problem is expounded.The multi-objective differential evolution algorithm is improved under unconstrained and constrained conditions.Under unconstrained conditions,the initial population is generated by the combination of chaotic mapping and reverse learning method,and the mutation and crossover operators are adjusted adaptively.The niche technology of clustering exclusion mechanism is used in fitness allocation.Under the constraint conditions,two improved methods are mainly studied.One is to carry out constrained processing according to the unconstrained improvement method,and improve the pareto environment selection strategy under constraints.Another improved method is to integrate with other algorithms.Choose to merge with genetic algorithm.The individual of pareto front of multi-objective genetic algorithm is replaced to the corresponding position in the initial population of multi-objective differential evolution algorithm.In the selection of parent and offspring,Cauchy perturbation is performed on some offspring populations,and then the improved pareto environment selection strategy in the first method is used to compare and select the offspring and the parent.Comparing the improved algorithm with other algorithms,the results show the superiority of the algorithm.(4)An improved multi-objective differential evolution algorithm using weighted vector is proposed and applied to reactive power optimization of distribution network.