采用堆叠长短期记忆神经网络的水质连续预测方法

Aiming at the issues of abnormal water quality parameters and low prediction accuracy in water environment monitoring, this paper proposes a water quality parameter prediction model based on stacked long short-term memory neural network (SLSTM) to tackle the challenge of incomplete time series data. First, the timing characteristics of missing or abnormal water quality data were analyzed, and a deep neural network model for water quality prediction was designed based on stacked long short-term memory networks. Second, point-by-point prediction and multistep prediction methods were used to validate the proposed model in comparative experiments. Lastly, in order to quantify the prediction performance of the model, two types of metrics were introduced, namely, the mean absolute percentage error (MAPE) and the root-mean-square error (RMSE) to assess the superiority of the SLSTM model over the support vector regression (SVR) and autoregressive integrated moving average (ARIMA) models. The experimental results showed that the prediction accuracy of SLSTM was significantly higher than that of the other two models in short-term (24 h) and long-term (48 h) chlorine residual prediction, the MAPE of SLSTM was at least 9. 15% lower than that of SVR for multistep prediction, and the RMSE of SLSTM was at least 31. 25% lower than that of SVR for point-by-point prediction. In addition, compared with the ARIMA model, SLSTM can capture the nonlinear trend of water quality data more effectively and improve the prediction stability. This study not only verifies the effectiveness of SLSTM in water quality parameter prediction, but also provides new perspectives and tools for the field of water environment monitoring.