Thermal performance prediction of energy-efficient roofs via a CNN-LSTM hybrid model optimized by the sparrow search algorithm

Building energy consumption accounts for a substantial share of overall societal energy use. As an effective measure to reduce building energy demand, energy-efficient roofs require rapid and accurate thermal-performance prediction to support building retrofit decisions. In this study, we integrate artificial intelligence with energy-efficient roof thermal-performance assessment and develop an SSA-optimized CNN-LSTM deep learning model. Using air temperature, relative humidity, solar radiation, and wind speed as inputs, we perform annual surface-temperature prediction and comparative analysis for nine roof modules formed by three energy-efficient roof types: ventilated insulation roofs (Air), water-storage roofs (Water), and green roofs (Green) and three thickness levels (10 cm, 20 cm, and 30 cm), with a bare roof as the baseline. The results reveal pronounced differences in predictive performance across roof types: the coefficients of determination R2 are 0.945/0.939/0.950 for Air 10/Air 20/Air30, 0.979/0.983/0.982 for Water10/Water20/Water30, and 0.959/0.957/0.960 for Green10/Green20/Green30. Overall, the Water series achieves the highest accuracy and stability, followed by the Green series, while the Air series performs relatively worse. A multi-metric scoring system provides actionable screening conclusions: Water20 and Water30 obtain the highest scores (both 86) and are identified as the best-performing modules, whereas Air20 and Air30 receive the lowest scores (25 and 23, respectively), indicating higher predictive uncertainty and the need for further improvement. Spearman rank-correlation analysis shows that air temperature has the strongest association with roof surface temperature (approximately 0.882-0.943), followed by solar radiation, while relative humidity and wind speed exhibit weaker correlations, indicating that air temperature is the dominant meteorological driver of roof thermal performance. These findings offer quantitative support for comparing and selecting energy-efficient roof schemes during the design and retrofit stages.