Water quality forecast through application of BP neural network at Sifangtai

To forecast the COD_Mn at Sifangtai of Songhua River, ANN technique was employed to establish a forecast model. It adopted LM (Levenberg-Marquardt) algorithm to achieve a higher convergent speed, and the early stop method to improve the extended capacity. Sample data of ANN were selected from daily measured values of Sifangtai Station from 1997 to 1999, and the Layida rule was adopted to eliminate abnormal data. To evaluate the veracity of this model, forecast effects of the abundant rain period, freezing period and other periods of Songhua River were investigated. And the forecast effects in recent and long terms after 1999 were compared. It is indicated that the forecast effect in abundant rain period is the worst, that in freezing period is the best and that in other periods is in the middle; the forecast effect in recent term is better than that in long term. The whole forecasting effects are good, and the model can be used for water quality management of Songhua River.