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

Abstract

The rapid expansion of distributed rooftop photovoltaic (PV) systems introduces increasing uncertainty in distribution grid planning, hosting capacity assessment, and voltage regulation. Reliable estimation of rooftop PV deployment from satellite imagery is therefore essential for accurate modeling of distributed generation at feeder and service-territory scales. However, conventional computer vision approaches rely on fixed learned representations and globally averaged visual correlations. This makes them sensitive to geographic distribution shifts caused by differences in roof materials, urban morphology, and imaging conditions across regions. To address these challenges, this paper proposes Solar Retrieval-Augmented Generation (Solar-RAG), a context-grounded framework for photovoltaic assessment that integrates similarity-based image retrieval with multimodal vision-language reasoning. Instead of producing predictions solely from internal model parameters, the proposed approach retrieves visually similar rooftop scenes with verified annotations and performs comparative reasoning against these examples during inference. This retrieval-guided mechanism provides geographically contextualized references that improve robustness under heterogeneous urban environments without requiring model retraining. The method outperform both conventional deep vision models and standalone vision-language models. Furthermore, feeder-level case studies show that improved PV inventory estimation reduces errors in voltage deviation analysis and hosting capacity assessment. The results demonstrate that the proposed method provides a scalable and geographically robust approach for monitoring distributed PV deployment. This enables more reliable integration of remote sensing data into distribution grid planning and distributed energy resource management.

Figures (15)

• Figure 1: Overview of the proposed Solar Retrieval-Augmented Generation (Solar-RAG) framework for photovoltaic assessment from satellite imagery. The framework performs retrieval-guided visual reasoning by combining satellite image embeddings with a globally distributed rooftop reference database. For each query image, visually similar rooftop scenes are retrieved and used as contextual evidence during inference. A vision–language model then performs comparative reasoning between the query rooftop and retrieved reference cases to estimate photovoltaic presence, panel quantity, and spatial location. By incorporating geographically relevant reference information into the inference process, the framework improves robustness under heterogeneous rooftop environments and enables scalable photovoltaic inventory estimation for distribution system analysis.
• Figure 2: Workflow for developing the rooftop reference repository used for distributed PV assessment. Candidate rooftop locations are first identified from open geospatial databases using the OpenStreetMap Overpass API. High-resolution overhead satellite imagery is then retrieved through the Google Maps API under standardized acquisition settings, forming a large rooftop image inventory. An automated attribute extraction module generates preliminary PV descriptors for each tile, including installation presence, panel quantity, spatial location, and contextual characteristics. These descriptors are subsequently reviewed and corrected through human validation to ensure reliability for planning applications. The validated rooftop tiles, together with their structured PV attributes, constitute the reference repository used for similarity-based case selection and reference-assisted PV estimation.
• Figure 3: Automatic labeling prompt used for structured PV descriptor extraction.
• Figure 4: Representative examples of similarity-based reference selection for rooftop PV assessment. For each input satellite image, the five most similar reference rooftop cases are identified using normalized numerical similarity measures. The similarity values reflect strong correspondence in rooftop geometry, panel appearance, and surrounding environmental conditions.
• Figure 5: Retrieval-augmented prompt structure used for PV assessment.