An Improved BP Algorithm Capable of Getting the Network Out of False Saturation Area

This paper analyses the existing conditions of false saturation in BP algorithm, gives the concept of the boundary area of network, and then proposes a simple and effective improved algorithm which can get the network out of false saturation area and speed up learning rate. The computer simulation results show that, with this improved algorithm, we can not only evidently speed up the convergence rate of learning, but also save the network from the exponential overflow in the nodes transformation function.