The Research On The Improvement Of Multi-objective Differential Evolutionary Algorithm

In academic research and engineering practice,there are many multi-objective optimization problem,different from the single objective optimization problems,multi-objective optimization problem because of the constraints between the various objectives,it is difficult to let all of the optimization target at the same time to achieve optimal.Therefore,it can only be coordinated with each objective to find the optimal solution.In addition,the multi objective optimization problem has many characteristics,such as high dimension,multi peak and discontinuity,so it is more difficult to solve the multi objective problem.Differential evolution(DE)algorithm as the most outstanding one of the intelligent optimization algorithms,with the characteristics of simple operation,less controllable parameters,fast convergence,global search ability strong,a large number of experimental data show that the difference evolution algorithm in handling the multi objectives optimization problems showed very significant effect,has been widely used in image processing,to mobilize the production,neural network,fault diagnosis in many fields.However,like other evolutionary algorithms,DE algorithm still has some problems,such as high dimension,multi peak,multi objective and so on.In this paper,the DE algorithm is studied and analyzed,and the contradiction between the exploration and development ability is analyzed.From the structure and key steps of algorithm,such as mutation,crossover operation of were in-depth analysis and a large number of simulation experiments,and finally puts forward two improved schemes,on the convergence speed and accuracy is greatly improved,in dealing with complex high dimensional multimodal problems have obvious improvement.The first scheme is presented in this paper,a DE algorithm based on classification strategy:the thought of classification,the populations were classified into several subgroups,to each sub group,according to the different characteristics of the different mutation strategies,in order to improve the convergence speed and precision of the algorithm.Specific improvement measures are as follows:1.Design a new DE mutation strategy DE/rand-to-best/pbest.To improve the convergence speed of the algorithm by using the guidance information which is provided by the past dynasties to improve the convergence rate of the algorithm;2.To adjust the degree of evolution of different characteristics of the individual,to balance the ability to explore and develop the algorithm.Experimental simulation results on 9 standard test functions show that the C-DE algorithm can effectively improve the convergence speed,accuracy and robustness,and its related performance indicators are better than the domestic and foreign advanced DE algorithm.The second scheme proposed a DE algorithm based on chaotic local search strategy:the chaotic local search strategy,randomly selected from the best individual of population in the fixed interval length as the local search center,in order to improve the algorithm performance.It is also divided into two steps:1.Design a new DE mutation strategy DE/best-to-pbest/1;2,the introduction of chaotic local search strategy.The convergence speed and the optimal value of the hybrid algorithm with local search are greatly accelerated.Experiments show that the C-DE and CL-DE algorithms have excellent performance,and the CL-DE algorithm is better in dealing with multi modal problems.Finally,the CL-DE algorithms are applied to the multi-objective optimization problem,and the corresponding multi-objective differential evolution algorithm CL-MODE is proposed.The simulation results show that the algorithm are superior in performance.