Multi-objective optimization of photovoltaic facades in prefabricated academic buildings using transfer learning and genetic algorithms

Abstract

Photovoltaic (PV) facades design in academic buildings requires balancing carbon emissions, daylighting, and thermal comfort. Traditional methods often enhance indoor comfort at the expense of higher carbon emissions. Thus, this study, leveraging transfer learning and genetic algorithms, integrates building simulation, performance prediction, optimization, and CFD analysis into a multi-objective optimization workflow. Targeting net-zero carbon emissions while maintaining indoor comfort, it optimizes classroom form, enclosure performance, fenestration, and PV shading devices. The results demonstrate that: (1) carbon emissions decrease by 27.88 kgCO₂/m², daylighting improves by 1.06 %, with thermal stability; (2) PV shading devices tilt angles, window-to-wall ratios, and classroom height significantly influence building performance; (3) the integration of LGBM, CNN, and NSGA-III effectively improves the efficiency of performance predictions and optimization; (4) recommended PV panel tilt angles (0–10°) and cavity depths (60 or 150 mm) effectively reduce facade surface temperatures and improve PV module efficiency. The findings provide a scientific basis for the extensive application of PV systems on prefabricated academic building facades.