Multi-objective optimization design of STPV curtain walls based on integrated machine learning

The design of semi-transparent photovoltaic curtain walls requires the simultaneous optimization of energy performance, daylight comfort, and economic feasibility, while overcoming the inefficiency of traditional simulation methods. This study proposes a machine learning-based multi-objective optimization framework that addresses this challenge. By integrating the NSGA-II algorithm with entropy-weighted TOPSIS decision-making, the framework achieves approximately 300-fold acceleration while maintaining solution quality, providing a climate-adaptive design tool for high-rise office buildings. A comparative analysis across five climate zones in China reveals differentiated optimal strategies: Beijing (cold region) selects high-transparency PV glass covering all areas except the north facade, balancing power generation and daylighting to achieve the highest Useful Daylight Illuminance improvement of 24.2 %; Shanghai (hot summer and cold winter) employs low-transparency, high-efficiency PV curtain walls on middle and high floors while avoiding low-rise and north-facing areas, resulting in the largest Energy Use Intensity reduction (12.93 kWh/m2) and shortest Dynamic Payback Period of 1.64 years; Guangzhou (hot summer and warm winter) adopts full-coverage installation to enhance shading, whereas Kunming (temperate) implements a zoned strategy with power generation on high floors and daylighting on low floors, with both outperforming baseline performance. The transferable configuration rules generated by this framework provide architects with practical, cross-climate design guidelines, facilitating the large-scale application of high-performance Building-Integrated Photovoltaic envelopes.