Energy saving and low carbon oriented renovation framework for educational buildings with Tianjin University case study.

Modern educational buildings often meet daylighting needs via higher window-to-wall ratios, yet this may cause excessive solar radiation and increased air-conditioning energy consumption. Amid global low - carbon and sustainable development, the energy efficiency and low-carbon design of educational buildings still lacks sufficient attention. This study proposes a data-driven workflow that combines a genetic algorithm with a Bayesian optimization-based XGBoost model to optimize the retrofit design of educational buildings. Taking the first educational building of Tianjin University as a case study, the study focuses on envelope material selection and shading measures, with optimization objectives including sustainable energy use (Av.LM), useful daylight illuminances (Av.UDI), and building carbon emissions (GWP). Building modeling, performance simulation, and multi-objective optimization were completed through Rhinoceros v8.0, Ladybug Tools v1.6.0 and WallaceiX v2.7 plugins, and data was processed using Python v3.10.0. The XGBoost model based on Bayesian optimization performed well in performance prediction with an accuracy of 0.86, precision of 0.77, recall of 0.86, and F1 score of 0.816, which is a significant advantage over LGBM, AdaBoost, and Random Forest models. Sensitivity analyses show that parameters such as north-facing window-to-wall ratio, facade and roof thicknesses significantly affect Av.LM and Av.UDI, while roof and facade selection and thicknesses have the greatest impact on GWP. The study's main contribution is a novel data - driven workflow balancing low - carbon goals and daylighting needs in educational buildings, validated through model comparison and sensitivity analyses. It highlights the potential to reduce energy consumption and carbon emissions while enhancing daylighting performance, aiding global low - carbon and sustainable development.