River flow prediction using artificial neural networks: self-adaptive error back-propagation algorithm

This paper presents a improved error back-propagation(BP) algorithm,which is named as self-adaptive error BP algorithm. The method includes two strategies:one is adding momentum term at the iterative expressions of the weights,the other is self-adaptive adjustment of the learning rate according to the variety of the sum error. Namely,if the sum error increases,the learning rate will decrease,conversely,the sum error decreases,the learning rate will increases. The learning rate is changed until the sum error has reached a satisfactory level. The improved algorithm can prevent the networks from getting in the part least and can shorten the study time. The self-adaptive error BP algorithm was utilized for predicting river flow of Yangtse River at Yichang station,and satisfactory results were acquired.