Enhancing the effluent prediction accuracy with insufficient data based on transfer learning and LSTM algorithm in WWTPs

Abstract

Recently, there has been rapid advancement in artificial intelligence (AI), revolutionizing numerous industries. But the application of AI in the wastewater treatment plants (WWTPs) is still in infancy with limited intelligence capabilities, due to insufficient data. In this study, a transfer learning algorithm based on a long and short-term memory neural network (TL-LSTM) was proposed for predicting effluent chemical oxygen demand (COD) in WWTP. The TL-LSTM approach involves the transfer of neural network parameters from a source domain with amounts of data to a target domain with insufficient data to improve prediction accuracy. Results indicate that TL-LSTM yielded a root mean square error (RMSE) of 0.627 mg/L and a determination coefficient R2 score of 0.811, outperforming a long and short-term memory neural network (LSTM) with RMSE of 0.754 mg/L and R2 of 0.727. Moreover, the effectiveness of TL-LSTM was further confirmed by other effluent indicators (NH3-N, TP and pH). Compared to LSTM, TL-LSTM achieves reduced RMSE by 16.8 %, 15.2 %, 7 % and 14.3 % for predicting effluent COD, NH3-N, TP and pH, respectively. Thus, this study underscores the relevance of TL-LSTM as a valuable tool for early warning, energy-saving and consumption reduction in WWTPs with insufficient data.