Urban expansion, driven by economic and technological development, has intensified building construction and increased energy demand. Over the last decade, this trend has become a global challenge due to the depletion of fossil fuel reserves and the environmental impacts associated with greenhouse gas emissions. As a result, integrating renewable energy sources has become essential for achieving sustainable development goals. Among the available alternatives, photovoltaic energy stands out as a solution to meet the growing energy demand in buildings. In this context, this study presents SolarisBIM.AI, a framework that integrates machine learning–based solar radiation forecasting with a Dynamo routine within a Building Information Modeling (BIM) environment to quantify solar energy produced and CO2 avoided. Three predictive approaches were tested: extreme gradient boosting (XGBoost), long short-term memory networks (LSTMs), and feedforward neural networks (FNNs). Using a 49-year historical solar radiation series from six cities in Pernambuco, Brazil, the FNN achieved the best performance. In a real case study, SolarisBIM.AI predicted an annual solar energy production of 381,802 kWh for 2023, only a 2% overestimation compared with the measured production of 374,204 kWh at the center for technology and geosciences (CTG) UFPE building. This close agreement reinforces the model’s suitability for early-stage design assessments and decision-making in building energy planning. These results demonstrate that the proposed approach can support architects, engineers, and planners in quantifying solar energy production and assessing building sustainability at the design stage.
Alves et al. (Tue,) studied this question.