Presentation Details
| A Minimal Neural Augmentation of PVLib for Accurate and Transferable PV Digital Twins Rafsan Sayad1, Shah Mohazzem Hossain1, Md Ahsan Kabir1, Mohammed Mynuddin2, Md Majedul Haque Mithun3, Mahir Foysal4. 1Military Institute of Science and Technology, Dhaka, Bangladesh.2Hitachi Energy, Raleigh, NC, USA.3North Carolina A & T State University, Greensboro, NC, USA.4University of Houston, Houston, TX, USA |
Abstract
Abstract — Existing models for estimating and predicting solar PV output are often computationally complex and require extensive per-site training, which limits their transferability and practical deployment. As such, this research presents a lightweight hybrid PV output power model that couples a PVLib physics core with a compact neural residual corrector to achieve accurate and transferable performance. The model deploys zero-shot across large variations in plant scale, climate, and tracking technology, immediately reducing PVLib only error by more than 50%. The hybrid preserves plausible behavior under extrapolation while substantially improving robustness to weather driven transients. The resulting architecture is data efficient, computationally minimal, and readily deployable, providing a practical foundation for scalable, fleet level PV digital twins without reliance on complex or over parameterized models.
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.
Content Locked. Log into a registered attendee account to access this presentation.