regression analysis;  artificial neural networks;  Water quality forecasting

Aiming at the outlier phenomenon in the prediction of artificial neural network, the prediction interval obtained by the regression analysis model is used to control the outlier phenomenon. Moreover, applied to the water quality prediction of the Sanmenxia section of the Yellow River, the average accuracy of the network model for ammonia nitrogen flux prediction before control was only 50.05%, because the predicted value in June 2006 deviated too much from the real value, and the relative error of the prediction reached 6.214%, which exceeded the regression prediction interval, thus affecting the overall accuracy. After control, the relative accuracy of the month was 88.90%, and the average accuracy reached 08.80%, and the overall prediction accuracy was significantly improved. Practice shows that this method is more effective for eliminating outliers in network model prediction.