Presentation Details
| Impact of data acquisition delays on photovoltaic forecasts performance in real-time operational processes Thomas Carriere, Alexandre Boilley, Florian Ghionda, Loic Yezeguelian, Sebastien Pitaval. Calibsun, Valbonne, France |
Abstract
With the growing penetration of photovoltaic (PV) generation in power systems, accurate PV power forecasting has become essential for grid operation and energy management. Most modern forecasting systems rely on a combination of Numerical Weather Predictions (NWP), satellite imagery, and in-situ measurements. While many studies report high forecast accuracy, these results are generally obtained under idealized, offline conditions and do not fully account for operational constraints encountered in real-time forecasting. In practice, several issues can degrade forecast performance, including faulty or missing measurements, delays in data acquisition, long computation times, and temporary unavailability of external data sources such as NWPs or satellite imagery. Quantifying the impact of these operational delays is therefore critical to obtain realistic performance benchmarks that better reflect real-world conditions. This work investigates how delays in satellite imagery and in-situ measurements affect PV power forecasting accuracy. The study relies on an extended Analog Ensemble (AnEn) model, which dynamically adapts to the set of available input features and can therefore handle delayed or missing data without retraining. Various delay scenarios were simulated and compared to an ideal reference case. Results show that for horizons below 30 minutes, measurement delays cause a sharp degradation in accuracy, with normalized Mean Absolute Error (nMAE) losses reaching up to −4.6% for a 30-minute delay and −6.2% when measurements are completely unavailable. Satellite data delays have a limited impact on short term forecasts when measurements are available but still induce roughly −1% loss in accuracy for horizons between 30 minutes and 3 hours.
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.