Research On Water Quality Prediction In Wastewater Treatment Plants Based On LSTM Neural Network

Wastewater treatment plant through purification technology for sewage treatment,the discharge of water can be used for agriculture,industry and municipal water,etc.The recycling of water has greatly alleviated the current situation of water scarcity.Wastewater treatment is a complex and variable process,susceptible to temperature,rainfall and other dynamic factors,with uncertainty,non-linearity,time delay and complexity,etc.Some water quality parameters cannot be obtained in real time due to their own characteristics,resulting in the control technology of the wastewater treatment plant cannot be adjusted accordingly in the first instance.In order to solve the problem that some of the key water quality parameters are not available in real time,some scholars use soft measurement technology to predict water quality,and achieved very good results.Artificial neural network has excellent fitting,is widely used in water quality prediction soft measurement modeling.Chemical oxygen demand（COD）,ammonia nitrogen（NH₃-N）and total phosphorus（TP）are selected as predictor variables in the original water quality index,and the prediction model of wastewater effluent quality is established based on long and short-term memory（LSTM）neural network,and a wastewater quality parameter visualization system is designed and implemented to provide reference for control technology optimization of wastewater treatment plants,while reducing economic costs and promoting the reuse of wastewater resources.The main research work in this paper includes the following points:（1）Selecting appropriate auxiliary variables for the three predictive variables of COD,NH₃-N and TP of the effluent to be predicted.Understanding the biochemical reaction principles of wastewater treatment processes related to effluent COD,NH₃-N and TP;selecting auxiliary variables with high correlation with the predictor variables;using PCA to reduce data dimensionality,remove data noise,redundancy and simplify the complexity of source data to obtain the auxiliary variables to be finally input into the prediction model.（2）A water quality prediction model using the LSTM network is established.Experiments using historical water quality data from the Lujiang City wastewater treatment plant,as the wastewater data has time series characteristics,and the LSTM neural network model can be a good solution to the problem of long time dependence that ordinary RNN networks cannot handle.The predicted values of effluent COD,effluent NH₃-N and effluent TP can be obtained by inputting the effluent data into the established LSTM model.Calculate the error between the predicted and actual values,obtain the MAPE,RMSE and MAE of the model,and judge the effectiveness and accuracy of the LSTM neural network model.The errors are calculated between the predicted and actual values to obtain the MAPE,RMSE and MAE of the model and to judge the effectiveness and accuracy of the model of LSTM l network.（3）A long and short-term memory network water quality prediction model（SSAA-LSTM）based on the Sparrow Search Algorithm（SSA）and the Attention mechanism（Attention）is proposed.The wastewater historical data is input into the model,the variation pattern within its features is learned through modelling,the attention mechanism is introduced,different weights are given to the implied states of the LSTM neural network through weighted mapping and learning parameter matrix,and finally the SSA optimization algorithm is applied to the selection of the model hyperparameters to improve the performance of the model.（4）High-latitude feature vectors are prone to the dimensional catastrophe problem.In order to reduce the influence of data dimensionality on the prediction results,a SSAA-LSTM model incorporating principal component analysis（PCA-SSAA-LSTM）,which performs PCA dimensionality reduction on the original data.The model without PCA dimensionality reduction was compared with the model with PCA dimensionality reduction,and the results showed that the PCA dimensionality reduction greatly improved the prediction accuracy of the model.（5）A critical water quality prediction system for wastewater was developed.In order to apply the PCA-SSAA-LSTM model to the actual wastewater treatment process,this paper designed a wastewater water quality prediction system.The system includes user login and registration module,key water quality parameters introduction module,data import module and water quality parameters prediction module.After the user logged into the system to call the relevant function module can be predicted on the water quality parameter value and get the water quality change curve,the user can understand the water quality situation accordingly.When the water parameters are not up to standard,the system can issue a timely warning to the user to ensure that the quality of treated sewage meets the national discharge standards and reduce the secondary pollution of the environment.