Differential Evolution Algorithm And Its Application

In optimization area, some common computation techniques such as Newton method, Conjugate Gradient Method, simplex are difficult to solve complex multimodal problems with high-dimension. For solving these problems, researchers given some advanced simulation algorithms to simulate them based on nature evolution progress. This common methods which come from the idea of nature evolution and we called them"evolution algorithm". These methods which don't need more specific information of problems adopt simple encoding technique to indicate complex structure, use simple processing on the encoding and selection method based on natural selection mechanism of the survival of the fittest to guide the search direction. Based on these special features, evolution algorithm are very simple and popular to use and get more efficiency easily.In recently years, a new evolution algorithm which called"Differential Evolution Algorithm"is very famous and popular. It has some features such as simple to use, fast convergence speed and need little information about problems. Through researching, we found it is very suitable to solve some complex optimization problems.Researchers began to research the algorithm from 2000 years and got a lot of production. Compared with other evolution algorithms, Differential evolution algorithm is more effective to solve optimization problems, but we found some faults and the algorithm is not mature. So we must keep on researching it and extend its application aspect.In this paper, firstly, we introduced the importance of differential evolution algorithm; secondly, we analyzed some special problems about differential evolution algorithm such as data structure, parameter setting, improved methods and application etc; thirdly, we used the algorithm to solve feature selection problem; lastly, we use it to optimize the neural network structure.The main contribution of this paper can be summarized as follows:(1) One the one hand, we fully introduced the theory, basic structure, implementing, improved methods and applications of differential evolution algorithm. On the other hand, we have done embedded research about parameters selection and setting problems.(2) Because differential evolution algorithm adopt real number encoding format, it is unsuitable to solve some discrete optimization problems. Then we addressed a binary encoding discrete differential evolution and use it to solve feature selection problems. Through experiment analysis, the method is efficient to get the best feature subset and improve the prediction veracity of classification algorithms.(3) The neural networks have many applications of engineering, especially for BP and RBF networks. But these networks are existed some faults such as lower convergence speed and easily get the local minimum etc. But differential evolution algorithm has some special features such as parallel computation, easily get global minimum and don't need continuity constraint. So, we use differential evolution algorithm to train BP and RBF networks structure and connection value, and then get evolutional network models. Through testing on datasets, these methods can get more efficiency.