Presentation Details
| Real-Time PV Power Nowcasting: A Two-Stage Hybrid CNN and LSSVM Framework Using All-Sky Imagers and Kalman Filtering (yes) Khadija Barhmi, Sara Mirbagheri Golroodbari1, Wilfried Van Sark. Utrecht University, Copernicus Institute of Sustainable Development, Utrecht, Netherlands |
Abstract
Accurate short-term photovoltaic (PV) power forecasting is essential for Building Energy Management Systems (BEMSs) and grid integration, enabling predictive battery dispatch, load scheduling, and optimized grid interaction. This study proposes a two-stage hybrid AI framework using one year of synchronized All-Sky Imager (ASI) images, PV power measurements, and local weather variables. In Stage 1, Global Horizontal Irradiance (GHI) is predicted for 1–30 minute horizons using CNN-based feature extraction combined with SVM classification and Kalman filtering for temporal consistency. In Stage 2, an LSSVM regression model maps forecasted GHI to panel-level PV power using meteorological inputs and time encodings. The framework was validated on campus-scale data from Utrecht, The Netherlands, achieving a normalized RMSE of 14.97–19.71% and reducing errors by 40–61% compared to persistence. These findings advance practical PV nowcasting by delivering an accurate, lightweight, and deployable solution that strengthens real-time BEMS control and improves grid flexibility through more reliable short-term generation awareness.
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.