Prediction of energy consumption and evaluation of affecting factors in a full-scale WWTP using a machine learning approach

Treatment of municipal wastewater to meet the stringent effluent quality standards is an energy-intensive process and the main contributor to the costs of wastewater treatment plants (WWTPs). Analysis and prediction of energy consumption (EC) are essential in designing and operating sustainable energy-saving WWTPs. In this study, the effect of wastewater, hydraulic, and climate-based parameters on the daily consumption of EC by East Melbourne WWTP was investigated based on the data collected over six years (2014−2019). Data engineering methods were applied to combine features from different resources. To this end, four various feature selection (FS) algorithms were used to reveal the relations among those variables and to select the most relevant variables for training the machine learning (ML) models. Further, the application of artificial neural networks (ANN) and two decision tree algorithms of Gradient Boosting Machine (GBM), and Random Forest (RF) were studied to predict EC records followed by a 95 % confidence interval assessment. Results of FS algorithms revealed that total nitrogen, chemical oxygen demand (COD), and inflow-flow had the highest impact on WWTP energy consumption. Moreover, GBM had the best performance prediction among all other regression algorithms. 95 % of confidence interval showed a reasonable prediction error band (±68 MWh/Day).