Dimensionality reduction and uncertainty quantification for high-dimensional sensor calibration in complex building energy systems

With the advancement of AI and big data technologies, the operation and maintenance of building energy systems have become more dependent on high-quality data. The high-dimensional in-situ calibration method integrating thermodynamic laws can calibrate sensor errors without removing existing sensors or adding redundant ones. However, practical applications often face challenges due to non-differential calibration processes and uncertainties inherent in solving high-dimensional optimization problems, which can compromise calibration accuracy. Addressing these challenges, this study introduces an innovative dimensionality reduction technique and an Uncertainty-Inclusive Basin Hopping (Un-BH) algorithm. Three case studies are conducted to demonstrate the effectiveness of the enhanced calibration method when applied to efficient chiller plant systems. The results demonstrate a substantial improvement in calibration results, with a significant decrease in Mean Absolute Percentage Error (MAPE) from 52 % to 1 % following dimensionality reduction. The impact of faulty sensor numbers and thresholds is also analyzed to validate the robustness of the proposed dimensionality reduction method. Furthermore, the uncertainty quantification of the Un-BH algorithm ensures reliable calibration results by avoiding convergence to local optima.