关键词: 自组织映射（SOM） 径向基函数（RBF） 隐层优化 聚类 噪声源识别

Abstract:

We present a new way which uses the Self-Organlzing Map（SOM） network to optimize the RBF hidden neurons to improve the velocity of identifying the a  coustic fault sources of underwater vehicles. Firstly, the former samples are clustered by SOM in the unsupervised form, then the weight between hidden  neurons and output neurons is adjusted with the BP（back-propagation） algorithm. Finally, we use new samples to test and compare the performance of the e original neural network classifier with the one after optimization. The experimental results show that it is an efficient method, and can improve the   velocity and accuracy of identification.

Key words: （self-organizing mapping （SOM）, radial basis function （RBF） ;optimization of hidden n eurons, clustering, a-coustic fault sources identification）