The Dynamical Analysis Of Evolutionary Algorithm

Abstract As a key technique of computational intelligence, evolutionary computation has attracted increasing interest because of its advantages of self-adaptation, parallelism, and robustness in solving complex and nonlinear problems. Although evolutionary computation has been widely applied to many fields, such as biological engineering, machine learning, self-adaptation control, neural networks, and economic prediction, these applications are often concerning special computational models only aimed at certain problems, and the general guiding principle is still in exploring stage. In the aspects of theory and applications of evolutionary computation, there are many problems need to be solved. Based on the review of recent development of evolutionary computation, this dissertation has studied and discussed the theory of evolutionary computation from dynamical analysis. It hopes to find a new way to researching evolutionary computation. The works of the dissertation are Based on UMDA.The works of the dissertation are described as followings:We proposed a UMDAe based on stochastic differential equation from improving traditional UMDA,estimated the potential function(based on Darwinian evolutionary theory) of stochastic differential equation to anlysis the dynamic behavior of UMDAe, proved the existence and uniquence of potential function ,and transform our problem into solving the (unstable)stable manifold,by numerical computation,we verified our potential function is accpetable.