Evaluation of the Effectiveness of Heat-Reflective Surface Treatment for Energy-Saving Building Envelope Based on Digital Twin

The heat flow in building walls is crucial to the air conditioning load of a building. Solar heat gain reduction is achieved by using heat reflective surfaces, such as retro-reflective paints. However, existing research lacks dynamic learning and predictive ability. This research proposes a digital twin model based on a transient heat transfer model. An Emperor Penguins Colony based Attention Recurrent Neural Network (EPC-Att-RNN) model is applied to improve the temperature, cooling load and energy consumption predictions. The RNN captures temporal correlations in thermal data, the attention mechanism highlights important time steps, and the EPC method calibrates the model parameters to improve convergence and performance. This research uses the Heat Reflective Building Energy Dataset, comprising 6,130 samples of environmental, building, operational, and thermal data. Imputation is first applied to handle missing data, normalization is then performed for feature scaling, and feature extraction is carried out to capture temporal patterns, thereby improving model reliability and prediction accuracy. The research was performed with Python 3.10. The results demonstrate that the model had good performance with a mean absolute error (MAE) of 15.6, root mean squared error (RMSE) of 24.498, mean squared error (MSE) of 600.122, coefficient of determination (R2) of 0.942, and mean absolute percentage error (MAPE) of 3.55%. At the 35th epoch, the lowest validation loss and the highest validation accuracy (0.888) were attained by the model. This holistic methodology provides a scalable intelligent way of improving the building envelope and designing energy-efficient buildings.