In view of the water quality prediction problem, taking the surface water quality monitoring factors as the research object, a Long Short-Term Memory (LSTM) neural network based model is proposed for water quality multi-factor prediction. At the same time, the proposed K-Similarity method is used to denoise the input data of the model to improve the prediction performance of the model. Compared with BP neural network, RNN, and traditional LSTM neural network prediction model, the experiment shows that the proposed method has the least square error and the prediction result is more accurate.