Abstract
Water quality and availability worldwide are greatly affected by climate changes due to global warming. For instance, recently, the water levels in numerous European rivers decreased compared to their levels in centuries. Treated and desalinated water represents a promising strategic option to mitigate water scarcity. The global cost of wastewater treatment plants (WWTPs) depends significantly on their energy consumption. A precise prediction of energy consumption of WWTPs can comprehend and predict the plant behavior to support process design and monitoring and enhance optimization of overall performances. This paper presents a practical machine-learning approach to predict WWTP energy consumption through Light Gradient Boosting Machine (LightGBM) approach. Data from a full-scale WWTP at Melbourne eastern over five years is employed in this study. Results indicate that the LightGBM is more accurate than the other fourteen machine commonly used machine-learning models. In addition, results revealed that including lagged measurements in constructing the investigated models improves the prediction accuracy. This study shows that the dynamic, optimized LightGBM model outperformed all models with reasonable root-mean-square error and mean absolute error of 37.38 and 28.63, respectively.