AI-empowered online optimization adopting local energy gradient for robust and energy-efficient control of building cooling systems

Building cooling systems account for a large portion of energy usage, making effective optimization crucial for improving building energy performance. Conventional optimization strategies rely on absolute energy consumption calculations, which are prone to instability and unreliability due to the influence of measurement data errors, thereby limiting their practical application. To address this limitation, this study introduces a local energy gradient-guided AI-empowered online optimization strategy that utilizes relative energy consumption in nearby regions to achieve optimization, thereby enhancing control stability and robustness to input data noise. This strategy involves theoretically formulating a physical variable, the “energy gradient”, defined as the rate of energy variation with respect to operational parameters. A machine learning model is employed to capture the trend of the energy gradient, guiding the progressive adjustment of setpoints. A comprehensive system mechanism analysis is conducted to uncover system dynamics and provide theoretical support for the proposed strategy. The proposed strategy is validated through hardware-in-the-loop tests on a physical control station. Results show that the proposed strategy exhibits superior stability and robustness, while achieving energy-saving of 5.23%–7.70% across different cooling load levels, comparable to conventional online optimization.