Presentation Details
| PV Copilot: A Large Language Model–Enabled Tool for End-to-End PV Degradation Analysis Baojie Li, Anubhav Jain. LAWRENCE BERKELEY NATIONAL LABORATORY, BERKELEY, CA, USA |
Abstract
Degradation analysis is critical for understanding long-term performance and reliability of photovoltaic (PV) systems. However, real-world PV data are often heterogeneous, inconsistently labeled, and require complex preprocessing pipelines that integrate multiple software tools. This paper presents PV Copilot, a free and open-access autonomous tool for end-to-end PV degradation analysis. It combines large language models (LLMs) with established PV analytics libraries to automate data ingestion, variable normalization, preprocessing, degradation computation, visualization, and code generation. We demonstrate the tool on 157 PV systems, achieving a data formatting accuracy exceeding 96%. Further validation on three PVDAQ field systems shows strong agreement between PV Copilot and expert-led manual analyses, with an R² score of 0.99. PV Copilot lowers the barrier to advanced analysis for users with limited coding experience and enables fast, extensible evaluation of PV system degradation.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.