Prediction of effluent COD of a sewage treatment plant in Tibet based on PCA-SSA-LSTM

Water quality prediction is the foundation of sewage prevention and control.The cold and high-altitude conditions in plateau areas have brought about a special environment of "three lows", namely low average temperature, low oxygen content and low air pressure. Temperature not only affects the metabolic activities of microbial populations and the oxygenation performance of equipment, but also influences the progress of various reactions during the sewage treatment process. Water quality data has characteristics such as nonlinearity, seasonality and regionality. This paper combines water quality data with temperature data, which can reflect the changing patterns of water quality and improve the prediction efficiency on the basis of ensuring the prediction accuracy. Too many indicators will increase the complexity of the model, and there is a certain correlation among the indicators. To seek a more accurate water quality prediction method, a water quality prediction model based on PCA-SSA-LSTM is proposed. The features of water quality influencing factors are extracted through PCA. The dimensionally reduced data is input into the LSTM for prediction, and the SSA is used to optimize the parameters. The prediction results of the three models, namely LSTM, PCA-LSTM and PCA-SSA-LSTM, were compared and analyzed. The results show that the PCA-SSA-LSTM model has the highest accuracy in predicting the effluent COD of sewage treatment plants, with R² reaching 0.9411, and RMSE, MAE and MAPE being 1.46, 1.18 and 3.78% respectively. This research provides important decision support for sewage treatment.