Building a three-layer BP neural network by fuzzy rules

The neural network has excellent learning ability, but the weights and the thresholds of it can´t be explained and understood. In this paper, a method to determine the structure and the parameters of a general three-layer BP neural network is proposed. The network is built based on the fuzzy rules obtained from the samples. The number of the nodes in the middle layer, and the parameters of the network can then be determined directly by the extracted fuzzy rules. The network constructed in this way can approximately track the output without learning. The feature of this method is that the structure and the traditional BP learning algorithm need not to be changed any more. Meanwhile, the significance of the weights and the thresholds can be explained with fuzzy rules. The simulation result proves the validity of the proposed method.