Presentation Details
| Evaluating Error Propagation Across the Photovoltaic Modeling Pipeline Through Blind Modeling Lelia Deville1, Kevin S Anderson2, Juergen Sutterlueti3, Youri Blom4, Mark Campanelli5, Alejandro Gonzalez Carballo6, Gregorio Olivares Casero7, Arijit Chakroborty8, Anton Driesse9, A S M Jahid Hasan`10, William B Hobbs11, Adam Ramsus Jensen12, Sha Li13, Andrei da Cunha Lima15, Ivan Diaz Martinez14, et al.1. 1University of Louisiana at Lafayette, Lafayette, LA, USA.2Sandia National Laboratories, Albuquerque, NM, USA.3Gantner Instruments, Schruns, Austria.4Delft University of Technology, Delft, Netherlands.5Intelligent Measurement Systems LLC, Bozeman, MT, USA.6EDP Renewables, Madrid, Spain.7CENER, Navarra, Spain.8Atkins Realis, Bengaluru, India.9PV Performance Labs, Freiburg, Germany.10North South University, Dhaka, Bangladesh.11Southern Company, Birmingham, AL, USA.12Technical University of Denmark, Kongens Lyngby, Denmark.13Leeward Renewable Energy, Dallas, TX, USA.14European Energy A/S, Søborg, Denmark.15Universidade Federal de Santa Maria, Santa Maria, Brazil |
Abstract
This blind modeling study organized by the PV Performance Modeling Collaborative (PVPMC) utilizes hourly and sub-hourly data taken from a lab-scale system in Albuquerque, NM and a utility-scale system in Germany to quantify how errors can accumulate throughout the modeling pipeline by modeling in iterative stages. At each stage, more measured information was provided to the participants. In the final stage, participants were asked to remove all loss assumptions. 31 distinct submissions from modelers from 11 different countries were submitted for analysis. The results of this study show that the effect of upstream models’ (i.e. plane-of-array [POA] and module temperature) errors directly affect downstream performance modeling. Module temperature errors were largely unaffected by errors in POA modeling (within 0.2℃). The power NRMSE in Albuquerque had a median decrease of 0.4% when provided the inputs and no derate guidelines whereas in Germany it increased by 1.5%. This underscores the need for improved modeling practices across all system types and sizes. The full study will be presented in a future manuscript with extended results and recommended best practices for modeling.
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.