Machine learning driven building integrated photovoltaic (BIPV) envelope design optimization

• BIPV envelope design optimization is a complex, time-consuming process. • ANN serves as an efficient surrogate model for BIPV performance prediction. • The NSGA-II algorithm effectively optimizes multiple BIPV envelope design objectives. • The proposed framework supports rapid BIPV envelope design optimization during the detailed design phase. • Computational time is significantly reduced by the surrogate-based MOO framework while balancing energy, cost, and comfort. By integrating renewable energy systems (RESs) into buildings, Building-Integrated Photovoltaics (BIPV) reshape the demand–supply relationship, offering substantial benefits such as reduced energy usage, lower greenhouse gas emissions, improved indoor comfort, and architectural enhancement. However, designing optimal BIPV systems involves balancing multiple parameters and conflicting objectives, making it a complex, computationally intensive process. This study proposes a data-driven optimization framework to enhance BIPV envelope design during the detailed design phase. The novelties of this research lie in 1) the use of Artificial Neural Networks (ANN) as a surrogate model for rapid prediction, and 2) a multi-objective optimization (MOO)framework tailored for BIPV design at the detailed design phase. The proposed framework integrates ANN-based predictive modelling with an NSGA-II optimization component to generate optimal BIPV configurations efficiently. Implemented in Python, this approach is validated through a benchmark case study in Melbourne, demonstrating its effectiveness in reducing computational demands while achieving balanced energy, economic, and indoor comfort performance. This research advances sustainable building design by addressing gaps in current optimization frameworks and providing practical tools for architects and designers, promoting the wider adoption of BIPV solutions with significant computational savings.