Large Scale Global Optimization Based On Cooperative Coevolution

With the development of engineering technology and the gradual improvement of the mathematical model of the optimization issue,many optimization problems develop from the initial low dimensional optimization to the present high dimensional optimization or the large-scale complex optimization.Therefore,the problem of large scale real-value optimization is a hot issue in scientific research,and it is also widely used in engineering practice.Although in recent years evolutionary optimization has achieved great success in many real-value optimization problems,most of the stochastic optimization algorithms(including particle swarm optimization,differential evolution algorithm and genetic algorithm)suffer "dimension disaster",which means that the performance of the algorithm is drastically degraded with the increase of the search space dimension.So compared with the low dimensional problems with simple topology,the global optimal solution of large-scale problem is difficult to find.To sum up,this thesis proposed a dynamic multi-swarm particle swarm optimizer with cooperative coevolution to solve large scale problem.The contents in this thesis include:First of all,the background and significance of the large-scale problem were introduced,also introduced the research status,development course and future hot research direction of the evolutionary algorithm in the field of large-scale optimization.This thesis focused on some basic concepts,basic ideas and characteristics of particle swarm optimization algorithm.And then the development course and research direction of particle swarm optimization algorithm in recent years were briefly described.At the same time,some solutions to large-scale optimization problems were introduced,which provided a basis for algorithm comparison.Next,the benchmark suits of large scale optimization algorithm were introduced,and the concrete functions of the test function set were listed.And the nature of the function and the characteristics of the large-scale problem were presented.Then,the initialization methods of large scale problem were introduced.By comparing with the initialization results of the basic random number generator,it was shown that the effects of different initialization methods were different in large-scale global optimization.The initialization method made the population of the algorithm distributed more evenly in the decision space,so that the algorithm could avoid falling into local optimal solution.Finally,the particle swarm optimization algorithm was used to solve the large scale problem.Aiming at the characteristics of large-scale problem,the dynamic multi-swarm cooperative co-evolution strategy was proposed,which focused on the characteristics of the new algorithm and the parameters involved in the algorithm.Also the mainly proposed algorithms in recent years were introduced.The effectiveness of the proposed dynamic multi-swarm particle swarm optimizer with cooperative coevolution is verified by the comparison experiments with these algorithms.