Abstract
In this paper, the structure and working principle of Radial Basis Function (RBF) Neural Network (NN) are analyzed. A new method for constructing and training of parallel RBF-NN is proposed. A compound heuristic Genetic Algorithm (GA) based on Singular Value Decomposition (SVD) is introduced for structure training of RBF-NN. Based on the advanced strategies proposed above, RBF-NN is used for fault diagnosis of turbo-generator. Computer simulation experimental results show that the approach is effective.