Cooperative Coevolutionary Algorithm Based On Correlation Analysis

Co-evolutionary algorithm is a simulation of cooperative evolution in ecology.The objects which take the cooperation involve multiple levels,including individual cooperation,population cooperation,and evaluation cooperation.Different levels will use different cooperative strategies.Cooperative co-evolutionary algorithm is to decompose a complex problem into subproblems.Cooperative evolution occurs between subproblems.In this cooperative mechanism,co-evolutionary algorithms make it more advantageous to solve complex large-scale problems.However,the decomposition of the problem in cooperative co-evolution is facing a new problem.Inappropriate grouping will lead to the destruction of the correlation between the original problem and affect the performance of the algorithm.In this paper,we analyze the correlation between different dimensions of the optimization problem and get the appropriate grouping of the problem.Under the guidance of this grouping,the problem is decomposed into several subproblems,and then the mutual optimal solution of the problem is obtained.By analyzing the correlation of existing optimization problems,a method of constructing test problem for binary is proposed.To solve the problem which is the correlation between dimensions of the test problem could not be directly measured,this paper indirectly reflects the correlation.The method proposed based on the concept of information theory,and by means of the joint entropy of different gene bits and function values to reflect the correlation.A new mertic method of correlation is used to evaluate the correlation between the constructed function dimensions based on Sample Desire Operator.The Clustering Operator is used to grouping the dimensions with correlation.The Cooperation Operator is used to make the sub-problems cooperate with each other.A cooperation co-evolution algorithm based on the correlation with those operators is proposed.The contents and achievements of this paper are described as follows:(1)The correlation measure method for binary coding and real number coding is compared and analyzed,and the correlation of some real number optimization problems is tested experimentally.(2)By analyzing the relationship between information entropy and correlation,it is found that the mutual information can represent the correlation between the loci.However the correlation could not be directly calculated.It is needed to use the corresponding function value of different genes to obtain the correlation between the loci.And a new method of correlation measurement is proposed.Some experiments are taken by measuring the optimization problems which are structured to be used in the binary population.And the results show that the correlation data obtained by the metric method is more resolvable.(3)Based on the new correlation measure,a new cooperative co-evolutionary algorithm based on correlation analysis is proposed.Under the action of the sample design operator and the clustering operator,the appropriate grouping of the problem in the evolutionary process is found.The subpopulation after the grouping is affected by the cooperative operator,and the excellent gene information is shared and transmitted.And through the experimental simulation,it proves that the new algorithm has obvious advantages in solving the problem of complex incomplete division.(4)A cellar genetic algorithm under multi-objective cooperation strategy is introduced in this paper.We construct an additional function to co-evaluate the object function in cell space,and design a cell evolutional strategy for multi-object problems.It can be improved in the experiments that the proposed algorithm can maintain diversity and avoid falling into local optimal solutions.