Research On Differential Evolution And Its Application

Evolutionary Computation (EC), as a novel intelligent optimization technology, has been utilized extensively in engineering science. It has more advantages over traditional optimization methods when solving problems with global optimization and solving complex problems. Differential Evolution (DE) algorithm is a kind of evolutionary algorithm based on population difference; it finds the solution of optimization problem through the cooperation and competition between individuals. Two subjects of research in DE are to expanse the application domain and to make it more effective. The former is for the goal of designing and discovering effective of DE search strategies, solving the problem that the past methods unable to be solved or cannot be solved effectively, and the latter is to revise and improve emphatically of the algorithm. Some novel methods are designed to improve the performance of DE with revolving the two subjects in the dissertation. In addition, some novel schemes are presented to solve some complex problems which we usual encounter with the proposed DE. The dissertation is organized as follows:Firstly, this thesis introduces the research background and significance; some branches of EC and its history are introduced briefly; then this paper introduces the principle of DE algorithm, development and research status. The paper focuses on the population initialization, the choice of parameters of differential evolution algorithm, strategy selection and hybrid algorithm. At last, the researching work and innovations points of the dissertation are briefly described.When focusing on the problem of differential evolution algorithm is easy to fall into local optimal solution, we put forward a new scale factor disturbance mechanism improvement. In order to improve the diversity of the population, the random distribution of scale factor vector instead of the value of the fixed scale factor is used in mutation phase.To obtaining the diversity of population, this paper puts forward using ’Partial Random Intermediate Recombination Crossover’ to improve the DE algorithm. During the crossover operation phase, through the use of local restructuring operations generated offspring among individual will get more information from the parent generation. This operation can obtain offspring within the hypercube defined by the characteristics of the dimensions of the target vector. In addition, the DE algorithm may appear stagnation; the proposed hybrid crossover operation can expand the search area in order to improve the global search ability of DE.When studying the selection of base vector of DE algorithm, this paper proposed the linear combination of best individuals and random individuals as the base vector. During the different stages of evolution, dynamic adjustment base vector in the proportion of individual best and random, both to the individual diversity of the population evolution of initial stage, and the convergence of the algorithm in the late stage of evolutionary are obtained.DE algorithm hybrids Group search optimization (GSO) method is proposed to improve the overall performance of the algorithm. The parameters of Adaptive resonance theory (ART) neural network need to be manually set beforehand, the DE algorithm is proposed to optimize the parameters of the SFAM. Experiment results of using the SFAM as a classifier show that the method can obtain higher classification accuracy.In order to eliminate the possible translation, rotation and scaling of the deviation of the Camera Space Manipulation (CSM), this thesis use DE to estimate the offset. The experimental results show that the CSM positioning precision has been obviously improved by using the DE algorithm.The thesis is summarized lastly, some problems existing in the DE algorithm are proposed and the possible future research areas of algorithm are prospected.