A study on Energy Saving Methods for Sewage Treatment Systems using Artificial Intelligence

This study proposes an artificial intelligence-based framework for reducing energy consumption in wastewater treatment plants while maintaining effluent quality. Three years of hourly data were used, including influent flow and quality, process operating variables, blower power consumption, and effluent organic matter, ammonium nitrogen, total nitrogen, and total phosphorus. After preprocessing, the data were chronologically divided into training, validation, and test sets at 70%, 15%, and 15%, respectively. An uncertainty-weighted multi-task TCN-LSTM model was developed to simultaneously predict blower power consumption and multiple effluent quality variables. The TCN extracted short- and medium-term temporal patterns, while the LSTM learned long-term dependencies and process delays. The trained model was then used as a surrogate model for constrained economic model predictive control. The dissolved oxygen setpoint, airflow rate, internal recycle ratio, and external recycle ratio were optimized under effluent quality, operating-range, and rate-of-change constraints. The NRMSE-based fit scores were 96.5%, 95.3%, and 95.2% for the training, validation, and test datasets, respectively. were 96.5%, 95.3%, and 95.2% for the training, validation, and test datasets, respectively. The proposed control strategy reduced relative energy consumption from 100% to 79.8%, indicating an estimated energy-saving potential of 20.2%, while satisfying all effluent quality constraints. These results indicate that the proposed framework can effectively support energy-efficient and stable wastewater treatment operation.