Research On River Water Quality Prediction Method Based On CEEMDAN-N-BEATS Combined Model

In order to predict the state of river water quality and the evolution trend of water quality,this paper proposes to use the CEEMDAN-N-BEATS algorithm based on time series to train the water quality prediction model.The water quality monitoring data of the Irtysh River in Altay,Xinjiang is used as a sample for training and verification to improve the integrity of the water quality data,linear interpolation is employed to fill in the missing values.Predict water quality indicators by predicting the corresponding water quality parameter components such as dissolved oxygen,ammonia nitrogen content,and potassium permanganate quality.Experimental results show that the CEEMDAN-N-BEATS water quality prediction model has high prediction accuracy,small relative errors,and reasonable prediction results.It can predict water quality indicators and water quality grades based on past water quality data and water quality evolution trends,which is water pollution Provide effective reference for prevention,protection of water quality and safety,and water supply safety.The primary work is summarized as follows:(1)The common methods and basic theories of water quality prediction are discussed.Incl uding regression analysis-based support vector regression and correlation vector regression model s,neural network-based back-propagation neural network and long-short-term memory network a nd other forecasting models,as well as time series based grey system correlation analysis and diff erential integration moving average autoregressive models.(2)The decomposition algorithm of water quality data is studied.The Empirical Mode Decomposition(EMD),Ensemble Empirical Mode Decomposition(EEMD)and Adaptive Complete Ensemble Empirical Mode Decomposition(CEEMDAN)are discussed respectively,and the related water quality decomposition experiments of the Irtysh River are carried out.The water quality data is decomposed by selecting an appropriate data decomposition algorithm.The results show that the CEEMDAN model has a better decomposition effect on the water quality data of the Irtysh River.Compared with the commonly used EMD and EEMD models in the past,it can better suppress the modal aliasing problem caused by data decomposition,and reduce modal splitting and reconstruction error.(3)The N-BEATS neural network is compared with the MLP neural network to verify the ability of a single model to predict the water quality level on the Irtysh River water quality data set.The model uses forward and backward residual links and a fully connected layer stack to construct the input variables of the model in a sliding window according to the time series sequence of the water quality data.The experimental results show that the N-BEATS water quality prediction model has good prediction results for solving the significant non-linear problem of the water quality data time series.It can better predict the water quality level of the Irtysh River within a certain period of time and save the evaluation of the water environment time.(4)The Irtysh River water quality prediction model based on the CEEMDAN-N-BEATS combined model is proposed and compared with the CEEMDAN-LSTM water quality prediction model of the Irtysh River based on the CEEMDAN-N-BEATS combination model was proposed,and compared with the CEEMDAN-LSTM water quality prediction model and the CEEMDANBP water quality prediction model.The experimental results show that from the perspective of water quality grade prediction,the CEEMDAN-N-BEATS prediction model and the CEEMDANLSTM water quality prediction model are better than the CEEMDAN-BP water quality prediction model in predicting the water quality of the Irtysh River.From the perspective of MSE,the CEEMDAN-N-BEATS water quality prediction model has a corresponding improvement in the prediction performance of the DO and CODMn values,which proves the excellent effect of the water quality prediction model in this paper.The monitoring and management of the surface water environment provide a certain reference value.