Post-construction evaluation of building envelope thermal performance using neural networks

The thermal performance of buildings is commonly evaluated only at the design stage, while post-construction evaluation has received limited attention and remains difficult to be compared under different climate conditions. This study proposes a data-driven methodology to evaluate the thermal transmittance based on indoor temperature variations during non-air-conditioned periods. A neural network (NN) model was developed to predict the indoor temperature difference between 0:00 and 6:00 (Tin,6-Tin,0) using the outdoor air temperature and initial indoor temperature at 0:00 (Tin,0). The model was validated by the measurements recorded by BACS in a university campus building. The effects of NN model configuration and the setting of Tin,0 on prediction accuracy were examined. The results demonstrate that the proposed model can accurately predict indoor temperature variations (Tin,6-Tin,0) during nighttime. Then, the method was applied to compare three rooms with different envelope thermal performance. Rooms with better envelope thermal performance exhibited smaller indoor temperature variations, whereas those with poorer performance showed larger temperature changes. Furthermore, the thermal behavior of these rooms was evaluated under different climatic regions using climate data. The results indicate that the performance differences between rooms become more pronounced in colder climates, highlighting the applicability of the proposed method for comparative envelope thermal performance evaluation across regions.