Improvement Of Multi-objective Distribution Estimation Algorithm And Its Application

There are many optimization problems in scientific research,engineering practice and social life,and most of these optimization problems are multi-objective optimization problems.With the development of evolutionary algorithms,people find that there are many advantages in solving multi-objective optimization problems by using evolutionary algorithms.Therefore,multi-objective evolutionary algorithm has become one of the hot topics in the field of multi-objective research.A regularity model-based multi-objective estimation of distribution algorithm(RM_MEDA)was proposed in 2008,which applied the distribution estimation algorithm to multi-objective optimization problems well.The simulation results also show that the algorithm has good performance.However,RM_MEDA also has some shortcomings,such as slow convergence speed,poor local search ability and premature phenomena.Based on RM_MEDA,this paper proposes an improved multi-objective estimation of distribution algorithm,the main research work are as follows:(1)Improve the elite strategy in RM_MEDA,set up an elite population to store the non-dominated solution generated of the father-son population in each evolution,and use the elite population as the next-generation parent population to guide population evolution in order to accelerate the population convergence;use adaptive niche technology to maintain the elite population,avoid premature,but also make the distribution of population more uniform;change the generation of offspring,using traditional crossover and mutation steps in NSGA-II when the non-dominated solutions are not enough,otherwise,using the estimation of distribution algorithm;add chaos partial optimization to the elite population in order to improve the accuracy of solution and the local search ability of the algorithm;(2)The simulation experiments were carried out on test functions with two targets and three targets respectively.The experimental results were compared with two classical multi-objective optimization algorithms,NSGA-II and RM_MEDA,and analysis the results in convergence metric,diversity metric,performance stability and the accuracy of solution.(3)On the vehicle routing problems with time windows,a reasonable model is given for the three objectives of the shortest distance,the least cost and the least vehicle.In the simulation experiment,a reasonable coding method is designed and the population isinitialized by the time difference insertion detection method.The simulation experiment is carried out on the test data of Solomon's VRPTW standard problem by using NSGA-II and the improved algorithm in this paper respectively,compare and analysis the results and algorithm performance.