Accurate detection and localization of photovoltaic (PV) panels from satellite imagery is essential for microgrid and active distribution network (ADN) optimization. Existing methods lack transparency regarding their algorithms or training datasets, rely on large, high-quality PV training data, and struggle to generalize to new geographic regions or diverse environmental conditions without extensive retraining. These limitations lead to inconsistent detection results, hindering large-scale deployment and data-driven grid optimization. In this paper, we explore a method to address these challenges by leveraging large-scale language models (LLMs). We propose a PV Assessment with LLMs (PVAL) framework, which involves task decomposition, output normalization, few-shot prompting, and fine-tuning using a curated, well-annotated PV dataset. PVAL minimizes computational overhead while ensuring transparency, scalability, and adaptability across heterogeneous datasets.
