Application of Data-Driven Models in Predicting Energy Intensity for Water Sources Allocation

This study explores data-driven models to predict energy intensity and optimize production allocation in Tampa Bay Water's system in Florida, which utilizes desalinated seawater, surface water, and groundwater as main water supply sources. Analyzing extensive data on water quality, chemical usage, production, and energy consumption revealed significant energy intensity variations: desalination consumed the most (13,240–14,340 kWh/MG), followed by groundwater (616–2450 kWh/MG, with Morris Bridge wellfield at 1901–2078 kWh/MG) and surface water (593.9–596.7 kWh/MG). Production volume was the primary determinant of energy intensity across all sources, with additional influences from temperature, total dissolved solids, and chemicals. Multiple machine learning algorithms were evaluated, with random forest performing best for desalination and XGBoost and linear regression showing moderate accuracy for surface water and groundwater, respectively. Two optimization approaches were proposed, namely linear programming and an iterative machine learning method. Though achieving similar optimal solutions, the linear method proved more computationally efficient.