Author(s): Songlin Han, Jibin Han

Linked Author(s): Songlin Han

Keywords: TDG; BP Neural Network; Spill; Multiple linear regression;

Abstract: To overcome the defeats of traditional data driven method in nonlinear relation, Back Propagation Neural Network (BPNN), as a data-driven method based on big data, has been applied to predict total dissolved gas saturation downstream of dams by using measured data. The measured hydrologic and water quality data during 1995-2018 at the Forebay and tailwater station of the John Day dam are collected from U.S. Army Corps of Engineers (USACE), which includes TDG, sensor depth, water temperature at sensor, barometric pressure, spill, total flow and water elevation. The TDG at the tailwater station is model output variable and the other seven variables at the forebay station are model input variables. Root mean squared error (RMSE), mean absolute error (MAE), coefficient of correlation (R) and Nash-Sutcliffe efficiency coefficient (NSE) are selected to evaluate the performances of the model. The different groups of the input variables for the model are compared to analyze the prediction accuracy. Meanwhile, the accuracy of the results is compared with the results of multiple linear regression (MLR) method. According to the results, it was found that the TDG at the tailwater station could be successfully estimated using the BPNN model and the model, which uses all the seven input variables is the best model among all others models.

DOI: https://doi.org/10.3850/38WC092019-0577

Year: 2019