Abstract: The traditional water quality prediction model is complicated in calculation and will cause errors in the case of large convection, so it is not applicable to intelligent water quality prediction in the era of big data. Based on the data of San Francisco Bay surface water quality research area, the authors studied the water quality of the research area by using data analysis, statistical testing, deep learning time series models,and other technical methods. Based on the principal component information, a long short term memory(LSTM) circulation neural network model was constructed, and further,the water quality of 5 surface water sampling sites was predicted. The results show that the long short term memory cyclic neural network model can effectively control the loop and memory unit structure through gates and the input characteristics of the incoming model, thereby reducing the complexity of the model. Moreover, the prediction accuracy of the two-layer long and short-term memory cyclic neural network model is 5.3 % higher than that of the single-layer long and short-term memory cyclic neural network model.

Key words: water quality prediction, deep learning, surface water, statistical test, data analysis