Machine Learning-based thermal environment classification and prediction for historical neighborhoods

With the protection and revitalization development, the outdoor thermal comfort of historical districts has gradually been paid attention by scholars. Previous studies have predominantly focused on qualitative typological analyses of street-morphological factor correlations using fixed paradigms. Yet systematic frameworks integrating quantitative classification, thermal performance assessment, and predictive typological modeling remain underdeveloped for historical districts.

This paper quantitatively classifies the thermal environment of historical districts and establishes a prediction model to guide design. Spearman analysis was used to examine correlations between thermal environment and morphological parameters across multiple dimensions. K-means clustering extracted typical street thermal environment characteristics from the database. A prediction model for block thermal environment types was then developed using the random forest algorithm with morphological parameters as input. The results show that street thermal environment correlations are significantly dependent on street geometry and exhibit marked spatiotemporal differences. Clustering identified three typical thermal environment types, and the machine learning model achieved an accuracy of 90.91%, outperforming empirical classification. This achievement has developed a toolchain for deducing the thermal environment design of historical districts by converting post-evaluation verification into a pre-design generation process, thereby providing robust support and guidance during the design stage.