A New Algorithm for Training an RBF Network

Artificial neural networks (ANNs) are important tools for function estimation. However, the existing ANNs for fitting functions at least have one of the following drawbacks: 1) low accuracy, 2) unstable training process, 3) long learning time, 4) too many hidden nodes. In this paper, based on the orthogonal least squares learning algorithm, a new approach is proposed which uses the gradient descent method to optimally determine the spread of RBFs for training an RBF network. The experimental results show the new method overcomes the above disadvantages even if fitting a chaotic signal