DOI: 10.29130/dubited.1902964 ISSN: 2148-2446

Electricity Generation Assessment of Photovoltaic Window System Using Data-Driven Modeling

Ayşegül Toprak
Show PDF Cite
Photovoltaic window systems have emerged as a promising building-integrated photovoltaic (BIPV) technology by enabling electricity generation while maintaining facade transparency. However, accurately estimating their electricity generation potential at early design stages remains challenging due to the complex influence of architectural geometry on solar exposure. In this study, a data-driven regression framework is developed to predict the total annual electricity generation of photovoltaic window systems using a parametric simulation dataset provided by the National Renewable Energy Laboratory (NREL). The proposed framework employs four architectural and geometric input parameters window-to-wall ratio, number of stories, floor height, and north–south to east–west aspect ratio which are typically available during the early stages of building design. Three regression models, namely Linear Regression, Random Forest, and Gradient Boosted Regression Trees, are systematically evaluated using a five-fold cross-validation strategy. Model performance is assessed using normalized error metrics, including root mean square error, mean absolute error, and the Pearson correlation coefficient. The results indicate that Gradient Boosted Regression Trees achieved the best predictive performance, with a normalized RMSE of 0.265, normalized MAE of 0.235, and Pearson correlation coefficient of 0.964. In the original physical scale, the corresponding RMSE and MAE were 545.67 kWh and 483.83 kWh, respectively. Feature importance analysis reveals that the number of stories is the most influential parameter governing photovoltaic window electricity generation, highlighting the dominant role of building vertical scale. Residual analysis further confirms the absence of systematic prediction bias across the range of electricity generation values. Overall, the proposed approach enables rapid and reliable estimation of photovoltaic window electricity generation using a limited set of early-stage architectural parameters. These findings demonstrate the potential of machine learning-based models as effective tools for data-driven performance assessment and decision support in PV-integrated building design.

More from our Archive