The paper combines artificial neural networks (ANNs) with Kalman filter real time adjustment technique in order to improve traditional ANNs model. The weights are trained by Kalman filter real time adjustment technique in the process of sample training, and then the weights are used for check.One case is flow forecasting for upper reach of Minjing River at Zipingpu station by using the method proposed in the paper, and the results are compared with single ANNs model and single Kalman filter model. The results show if Kalman filter technique is used in estimating networks weights,hydrologic forecast accuracy may be improved.
