Enhancing in-situ model accuracy in operational building systems with augmentation-based synthetic operational data

Developing models in intelligent building systems is crucial for implementing advanced control, monitoring, and holistic maintenance of building systems. These models are constructed based on the data generated by the system. However, data acquired in real-world scenarios can have practical limitations that hinder achieving acceptable model performance. Real data may have quantitative limitations that may be insufficient in quantity depending on the measurement period or measurement time of the data, and qualitative limitations that are unbalanced or insufficient information in the data. Against this backdrop, this study proposes a data augmentation method to improve the performance of models constructed in the real system. The novelty of this method lies in spatially transforming the data to perform augmentation. The proposed approach has the advantage of performing augmentation using only data, without requiring additional simulations or reference systems for the augmentation process. The effectiveness of this method was discussed by applying it to the real building energy systems. The results indicate that the proposed approach shows significant benefits when the modeling source data is insufficient (less than two weeks) or relatively less informative.