Retrieval-Guided Photovoltaic Inventory Estimation from Satellite Imagery for Distribution Grid Planning

A team of researchers from Arizona State University and other institutions has published a novel AI framework called Solar Retrieval-Augmented Generation (Solar-RAG) that addresses a critical challenge in energy infrastructure planning. The rapid, decentralized expansion of rooftop photovoltaic (PV) systems creates significant uncertainty for utility companies trying to manage grid stability, voltage regulation, and hosting capacity. Traditional computer vision models, which rely on fixed learned representations, often fail when faced with geographic distribution shifts caused by varying roof materials, urban layouts, and imaging conditions.

Solar-RAG tackles this by integrating similarity-based image retrieval with multimodal vision-language reasoning. Instead of making predictions based solely on its internal parameters, the system first retrieves visually similar rooftop scenes that have verified annotations. It then performs comparative reasoning against these real-world examples during inference. This retrieval-guided mechanism provides geographically contextualized references, making the system far more robust across heterogeneous urban environments without the need for costly and time-consuming model retraining for each new region.

The paper demonstrates that Solar-RAG outperforms both conventional deep vision models and standalone vision-language models. Crucially, feeder-level case studies show that the improved PV inventory estimation directly translates to reduced errors in critical grid planning metrics like voltage deviation analysis and hosting capacity assessment. This provides a scalable and geographically robust method for utilities to monitor distributed PV deployment accurately, enabling more reliable integration of remote sensing data into distribution grid planning and distributed energy resource (DER) management strategies.