Fuzzy parameter adaptation for error backpropagation algorithm

The error backpropagation (EBP) learning algorithm has been widely used to train the feedforward artificial neural networks (ANN) in many practical applications. Due to slow convergence of this learning scheme, some changes have been reported in the literate in order to overcome this shortcoming. However, almost all of them are not robust enough, since not all the parameters related to the training speed were considered. Therefore, in this paper, a new learning scheme which consists of the standard EBP learning algorithm and fuzzy modeling is proposed. The fuzzy modeling, which is able to dynamically adjust the standard EBP parameters including learning rate, momentum, and steepness of activation function, is employed to speed up the learning speed. The proposed learning scheme is developed and implemented in C language. The simulation results demonstrate that it is able to solve the problem of slow convergence and more suitable than the standard EBP learning algorithm for the practical applications.