An improved RBF neural network with the adaptive spread coefficient

It is known the radial base function neural network (RBF-NN) is much efficient on the fitting or approximating for complex signals. The spread coefficient (Sc) is one of important parameters in the RBF learning algorithm. A suitable Sc can speed up the signal fitting process. This paper presents an improved RBF learning method based on the adaptive spread coefficient for the signal approximation of complex models. An actual signal approximating of a nonlinear model is applied to validate the effects of algorithm. The simulations show the presented RBF-NN has good effects on speeding up the learning and approximating performance, especially suitable for the real-time request of complex system modeling. The algorithm also shows an excellent performance on learning convergence.