An ensemble learning framework for rooftop photovoltaic project site selection

The selection of suitable locations for rooftop photovoltaic projects (RPVP) is critical for optimizing power generation efficiency and return on investment. However, traditional methods of site selection that rely on subjective assessments of index weights can compromise accuracy, while complex calculations may limit adaptability to changing real-world data. In this study, we proposed a data-driven ensemble learning framework that integrates socio-economic, environmental, climate, and geography factors to optimize RPVP site selection. Using data from 1589 counties in China, we mapped eight criteria to feature variables to facilitate machine learning classification. Furthermore, the K-means algorithm was employed to enhance the model's robustness against outliers. The findings indicate that the proposed stacking model exhibits superior performance in comparison to other classifiers, as evidenced by the higher scores of performance metrics. Specifically, for positive instance prediction, the stacking model achieves the highest Precision scores. According to the rankings of Precision scores derived from the four ensembled models, we categorized counties suitable for RPVP development into five priority tiers. The ensemble learning framework provides a valuable and reusable tool for advancing county-level RPVP site selection and serves as a motivation for selecting other renewable power plant sites.