Building Digital Twins for Thermal Pseudo-Measurements Generation

Due to practical constrains performing temperature measurements at user-relevant locations, limited insights about system states are provided to system operators, which constitutes a challenge in building monitoring and control. This paper proposes a digital twin architecture grounded in correlation-based Gaussian processes that provide thermal information in the form of pseudo-measurements. Tested with thermal measurements collected from the field, close-to-person pseudo-temperatures are provided based on the continuous input of remotely located temperature measurement signals. In particular, we show that the proposed approach is trainable on only a few days of measurements, which rules out neural network architectures, and outperforms alternative algorithms. On top, the results indicate a better performance when operated on multiple reference signals.