Data-driven modeling and bayesian optimization of cooling towers for the reduction of water consumption

Abstract

This study presents a data-driven framework that integrates Machine Learning and Bayesian Optimization to minimize water consumption in industrial cooling towers while preserving cooling efficiency. Using historical operational and environmental data from a power generation facility, several regression models (Linear Regression, Random Forest, XGBoost, and Neural Networks) were developed to predict makeup water flow. Random Forest and XGBoost achieved the highest accuracy, with R2 scores of 0.982 and 0.972, respectively. Bayesian Optimization was employed to efficiently tune hyperparameters, yielding substantial improvements in predictive performance such as reducing RMSE by up to 18.6%. The methodology also incorporated feature importance analysis, which identified critical operational drivers such as blowdown flow and inlet water temperature. Overall, Random Forest was preferred due to its superior predictive accuracy, ease of interpretation, and practical integration into operational dashboards. By combining predictive modeling, optimization, and interpretability, the study offers a powerful methodology for a data-driven tool to support decision-making and identify opportunities for minimizing makeup water use in cooling tower operation.