Multi-objective design optimization of educational buildings using CatBoost and NSGA-III under future climate scenarios

In the context of accelerating climate change and rising energy challenges, enhancing the adaptability and sustainability of educational buildings is crucial. This study proposes a framework integrating multi-source data, climate scenario simulations, surrogate modeling, and multi-objective optimization to overcome the limitations of conventional building performance optimization approaches in handling multiple building types, future climate scenarios, and mixed design variables. Hourly weather data for 2050 and 2080 under four CMIP6-based SSP scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8.5) were used to simulate the thermal, daylighting, and energy performance of three typical school building types (I-, L-, and C-shaped) in a hot-summer and cold-winter climate. Six performance indicators were considered, including daylight performance metrics, building energy use intensity, and thermal comfort assessment indices. The CatBoost-based surrogate model accelerated simulations by approximately 180 times and achieved higher accuracy with discrete variables. Integrated training outperformed individual models and enabled the simultaneous simulation of different building types. In multi-scenario optimization, fixing gene positions for climate scenarios and building typologies within the NSGA chromosome structure allowed the model to better capture cross-scenario interactions and accelerate convergence. Window-to-wall ratio, orientation, shading configuration, and wall material were key determinants of energy and thermal adaptability, while greater room depth improved comfort and reduced energy use. Optimal solutions favored south-southeast orientations and thicker exterior walls. This research provides a scalable, interpretable, and adaptive method for sustainable building design under climate uncertainty.