Abstract:

The effluent total nitrogen (TN) is one of the key indicators for assessing the biological denitrification performance of wastewater treatment plants(WWTPs). To mitigate the prevalent issue of excessive TN discharges from WTTPs, we proposed a real-time prediction model based on long short-term memory (LSTM) networks. We performed Pearson correlation analysis to determine model inputs and used grid search algorithm to optimize model hyperparameters. Then, we used the proposed model to predict the actual effluent TN in a WWTP in Chongqing and compared its predictive performance with that of traditional time-series models. Results indicate that the proposed model can effectively predict effluent TN with an average absolute error of 0.911 mg/L, an average root mean square error of 1.074 mg/L, and an average absolute percentage error of 11.28%. All of these performance indicators surpass those of the recurrent neural network and ARIMA models. The proposed model can serve as the foundation for effective monitoring of effluent TN.

Key words: long short-term memory model, Pearson correlation, grid search algorithm, effluent total nitrogen