Graphical and interpretable control guidance for real-life large cooling water system

Advanced model-predictive and reinforcement learning controls have demonstrated potential in building energy efficiency. However, limited interpretability of such controls for facility managers hinder their widespread use. This study proposes a graphical and interpretable control guidance for facility operators. The target real-life factory building is located in Busan, South Korea, being served by five chillers, five chilled water pumps, six cooling water pumps, and seven cooling towers. First, the authors developed hybrid models for the chillers and cooling towers by combining physical principles, specifications, and measured data. Next, the authors visually presented the total power according to feasible control options (including the number of chillers, chilled water pumps, cooling water pumps, and cooling tower fans) while meeting cooling demand and operational constraints. The developed models demonstrated sufficient predictive performance, achieving an MAE of 0.04–0.35 for the five chiller’s COP and 0.53 °C for the cooling towers’ outlet temperature. The authors compared the graphical results at three time points—two with similar cooling loads and one under a high cooling load. The visually presented control guidance displays multiple feasible control options, providing efficient strategies based on operational and outdoor conditions as well as enhancing interpretability for facility managers.