A novel in-situ sensor calibration method for building thermal systems based on virtual samples and autoencoder
Zhe Sun,Qiwei Yao,Huaqiang Jin,Yingjie Xu,Wei Hang,Hongyu Chen,Kang Li,Ling Shi,Jiangping Gu,Qinjian Zhang,Xi Shen
DOI: https://doi.org/10.1016/j.energy.2024.131314
IF: 9
2024-04-20
Energy
Abstract:Sensor networks are playing an increasingly important role in modern buildings. With the growing size of building sensor networks and the increasing use of low-cost sensors, the accuracy and reliability of these sensor networks face challenges. Therefore, in-situ calibration of sensor networks is crucial to maintain data quality. Various state-of-the-art methods typically require meeting stringent conditions, such as reference sensors or co-located sensors, accurate physical models, and a large amount of operational data, limiting their applicability in some scenarios. This paper addresses a common issue in sensor calibration: the non-differential calibration issue in uncontrolled environments. We propose an in-situ calibration method based on virtual samples and Autoencoder. Virtual samples are generated through Monte Carlo sampling to ensure the completeness of sample information. Autoencoder autonomously establishes relationships within the sensor network, integrating sensor fault detection and calibration into one step. Offline experiments optimize methods, and online experiments are utilized for verification and analysis. The online experiments demonstrated that the proposed method achieved a calibration accuracy above 98.9 % for single and multiple sensor faults, with a calibration error below 3 %. Compared to the SoT methods, our approach consistently delivered superior performance, confirming its outstanding efficacy.
energy & fuels,thermodynamics