Research On Watershed Water Environment Quality Prediction Based On Coupling Of Data-driven Model And Mechanism Model

As the economy develops and the process of industrialization accelerates,facing more and more prominent water safety issues in China.In the prevention and control of water resources pollution,efficient and accurate water quality predictions in the basin can clearly reflect the changing trend of water quality in the Watershed.This can adjust various water resources protection measures in time,transform water pollution from ex-post treatment to pre-prevention,and effectively improve the current situation of water pollution prevention and control.In recent years,with the establishment of more and more water quality prediction systems,water quality prediction technology has become perfect.However,in the face of complex water quality changes,it is difficult for the existing water quality prediction models to achieve stable and accurate predictions.It is urgent to use different models Coupling and combining the advantages of each model to improve the prediction effect of complex water quality problems.This article summarizes the domestic and international watershed water environmental quality prediction research theories,selects typical water quality indicators for the change characteristics of the water quality of the research section,and establishes a combination of data-driven models and a coupling of data-driven models and mechanism models.Carrying out research on the prediction of water environment quality in the basin,the main research work and results are as follows:(1)In order to improve the accuracy and stability of the data-driven model and enhance the universality of the model,a water quality prediction method based on neural network dynamic optimization combination weights is proposed.Aiming at the seasonality and trend of the water quality change of the study section,the SARIMA model considering seasonality and the ARIMA model without seasonality are established,which are determined according to ACF,PACF,AIC,seasonal factors etc.The optimal parameters are obtained,and the verification results show that the SARIMA model is obviously better than the ARIMA model,and the SARIMA model can predict the changing trend of water quality very well.Aiming at the multiple correlations between water quality indicators,an LSTM model is established,and the relevant indicators of the research section dissolved oxygen,permanganate index,ammonia nitrogen and total phosphorus are used as input values,and the number of hidden layers and the number of network iterations are obtained by using the trial algorithm.The optimal model parameter values.The verification results show that the average relative error rates of the predictions of dissolved oxygen,permanganate index,ammonia nitrogen and total phosphorus are 2.97%,2.29%,9.94% and 1.81%,and the prediction effect is good.The LSTM model is used as a dynamic optimization combiner,and the combined model LSTM-LSTM-SARIMA is obtained by training with a large number of LSTM models and SARIMA model prediction values from the research section.The verification results of the three models are compared and found: the average relative error rate of the combined model Below 10%,the prediction effect is significantly better than a single model.(2)Aiming at the sudden characteristics of water quality changes,a water quality early warning and forecasting method coupled with a data-driven model and a mechanism model is proposed.The mechanism model was established using MIKE11,and the parameters of the model were calibrated to obtain a roughness of the bottom bed of the reach of 0.04,and value of a decay coefficient k,ammonia nitrogen,and total phosphorus were 0.08 1/d,0.085 1/d,0.04171/d.The result of model verification shows that the relative error of 30-day prediction of permanganate index,ammonia nitrogen and total phosphorus is below 30%,and the established mechanism model has certain applicability in this research area.In order to further improve the prediction of sudden characteristic water quality,the data-driven model is coupled with the mechanism model: the data-driven model realizes the prediction of hydrological and water quality data,the predicted hydrological data is used as the boundary condition of the mechanism model,and the mechanism model is used to carry out sudden accidents.The scenario is set to realize the prediction of the increase of the downstream pollutant concentration after the sudden pollution event,and the permanganate index and the ammonia nitrogen pollutant concentration change curve of the section B and the section C are analyzed.The combination model LSTM-LSTM-SARIMA,which is based on the neural network to dynamically optimize the combination of weights,has good prediction results for water quality with seasonal,trend,and multiple correlation characteristics.The data-driven model and the mechanism model coupling model can be used for sudden pollution events Then carry out water quality prediction.In view of different water quality characteristics,corresponding stable and accurate prediction methods are provided,which provide reference for future water quality prediction.