Abstract: Water quality prediction is an important part of water pollution prevention and control. To improve the accuracy of water quality prediction and the early warning mechanism of the Sanmenxia Reservoir on the mainstream of the Yellow River, five reservoir water quality monitoring indicators including pH, dissolved oxygen, ammonia nitrogen, total phosphorus and permanganate index, were selected for accurate prediction of water quality. The VMD-SSA-LSTM model and VMD-SSA-SVR model were constructed. The variational mode decomposition (VMD) was used to denoise the river water quality data. The sparrow search algorithm (SSA) was selected to optimize the model parameters of long short-term memory artificial neural network (LSTM) and support vector regression (SVR), and the prediction effects of the two models were compared by mean absolute error (MAE), mean bias error (MBE), mean square error (MSE) and root mean square error (RMSE). The results showed that the four error index values of the prediction results of the VMD-SSA-SVR model were smaller than that of the VMD-SSA-LSTM model, indicating that the VMD-SSA-SVR model had a more accurate prediction effect on river water quality changes. This study enriches the research on the river water quality prediction model in the Sanmenxia section of the Yellow River Basin, and provides technical reference for ecological protection and high-quality development of the Yellow River Basin.