Rbf Neural Network Research And Applications

RBF neural network is a high effective neural network. It has the best and universalapproximation property, simple structure and fast training speed. In addition, this network modelcan be widely applied to the fields such as pattern recognition, nonlinear function approximation.The main works in the thesis can be stated as follows:1 A new kind of optimized algorithm is proposed, which is the Micro-BCC algorithm based on theBacterial Colony Chemotaxis optimized algorithm. Micro-BCC algorithm takes advantages of twobacterial colonies(optimizing bacterial colony and storage bacterial colony) to find the optimum.Optimizing bacterial colony is to find the optimum by BCC algorithm and storage bacterial colonyis to ensure the variety of optimizing bacterial colony. Compared with SGA and BCC algorithm,the M-BCCA is better.2 The Hybrid Structure Optimization Algorithm of Radial Basis Function Neural Networks basedon the former algorithm is proposed. In the Hybrid Structure Optimization Algorithm of RBFNN,the stop condition for Recursion Orthogonal Least Square (ROLS) algorithm is improved, andthe optimal number of hidden neurons in RBFNN is chosen by this improved ROLS algorithm;The thought of Bacterial Colony Chemotaxis algorithm is applied to determine the controllingparameters of the hidden neurons in RBF neural network. The improved ROLS algorithm iscombined with the BCC algorithm, i.e., ROLS-MBCC algorithm, so as to optimize the overallstructure of RBFN (including obtaining the appropriate structure and the appropriate controllingparameters). It's better than traditional algorithm, which is K-means and ROLS algorithm, in theefficiency and generalization.3 In the last part, this article uses the RBFNN to recognize mobile sign, uses the non-entirecharacter input and the multi-layered recognizer in the recognition, the RBFNN algorithm uses thehybrid structure optimization algorithm. The simulation experiment indicated that, compared withthe car license recognition with the entire character input, this method has big superiority in thetime order of complexity, and compared with the mobile sign recognition based on K-meansalgorithm, this method has certain superiority in generalization ability.