A New Method for Decision on the Structure of RBF Neural Network

RBF, as a feedforward neural network with single hidden layer, is applied widely in signal disposing, system modeling, control fields, etc. But the decision of its structure lacks effective methods. The discussion on ability of network generalization ability is one of important research aspects. The paper proposed a method based on PCA to decide the number of hidden neurons. Firstly it gives the larger number of network hidden neurons and compute the output of hidden layer, then makes PCA on it, calculates the cumulative explained variance rate and gets the number of principal components as the number of hidden neurons. The method has certain optimization ability to confirm the structure, which not only simplifies the generalization ability, but also has robustness to noises