Rbf Neural Network Based On Ant Colony Optimization Algorithm

RBF neural network has been successfully used in many fields,because of its capability of simple structure,fast training speed and good generalization ability.The choices of centers of hidden layer and the corresponding widths are very important and directly affect the goodness of fit of overall network approximation capabilities.Ant colony optimization(ACO) was introduced by M.Dorigo and colleagues as a novel nature- inspired heuristic based on the phenomenon of real ants foraging behavior in the early 1990s,it has the good ability to find the better answers.The paper mainly studies how to use the ant colony optimization algorithm to optimize the center and the width of hidden layer of RBFNN.The main works in the paper can be stated as follows:1.ACO has become an independent branch of intelligent computation and has been discussed as a special session in many international conferences.More and more researchers have paid attention to ACO.The paper mainly describes the mathematic description and features of the ant colony algorithm, introduces some modified algorithms,the development and application situation of the ant colony algorithm.2.Based on the feature of parallel search optimum of the ant colony algorithm and a dynamic method to adjust the parameter of evaporation coefficient,the center of each basis function of RBF can be defined by using a new proposed clustering algorithm.Meanwhile,in order to simplify the structure of RBFNN,we use a pruning method to remove those hidden units which make insignificant contribution to the overall network output.The RBFNN optimized by the optimization algorithm has a smaller structure,and the generalization ability of RBFNN is improved.3.Ant colony algorithm is improved and the parameters of RBFNN are optimized in two phases.The first phase can be described as:First,the network parameters are optimized by the modified ant colony algorithm;second,the parameters optimized in the first phase are further optimized using the steepest descent algorithm in order to get more acute neural parameters.We apply the RBF neural networks optimized by this algorithm in two classification problems and face recognition,and the experiments indicate that the calculation speed is fast and the generation ability of RBFNN is improved.