Multi-objective optimization of shading devices for glass corridors in historic building extensions: an interpretable machine learning based framework

Glass corridors are a common solution in the renovation and expansion of historic buildings, and improving the internal daylight and the thermal environment is crucial to the comfort of indoor occupants. As a prevalent passive design strategy, facade shading devices are widely applied in regulating the performance of glass corridors. Taking the glass corridor added to a university library in eastern China as a case study, this research explores the optimization effect of facade shading devices on the performance of glass corridors by combining machine learning with a multi-objective optimization algorithm. The results show that: (1) A machine learning framework is developed for optimizing glass corridors in historic buildings, with key design variables identified via interpretable analysis; (2) In the comparison of c algorithm performance, Adaboost achieves better predictive performance for Useful Daylight Illuminance (UDI) and Predicted Mean Vote (PMV), while Extra-trees exhibits a superior predictive performance for Daylight Autonomy (DA) and Daylight Uniformity (U); (3) Through multi-objective optimization, the optimal solution achieves a UDI of 63.68%, reduces PMV to 0.15, and maintains a U-value above 0.6. Its daylight, thermal and energy performance outperforms conventional shading across different climate zones. This study provides valuable theoretical and practical support for the sustainable design of historic building renovation and extension, helping balance architectural context protection, space utilization and low-carbon performance optimization.