Title:	Convolutional Neural Networks Applied to Sky Images for Short-Term Solar Irradiance Forecasting
Author(s):	Q. Paletta, J. Lasenby
Keywords:	Grid Management, Irradiance Forecast, Information and Communication Technologies, Computer Vision, Deep Learning
Topic:	PV Applications and Integration
Subtopic:	PV Driven Energy Management and System Integration
Event:	37th European Photovoltaic Solar Energy Conference and Exhibition
Session:	6BV.5.15
Pages:	1834 - 1837
ISBN:	3-936338-73-6
Paper DOI:	10.4229/EUPVSEC20202020-6BV.5.15
Price:	0,00 EUR
Document(s):	paper, poster

Abstract/Summary:

---

Despite considerable advances in the estimation of the solar resource, there is still a need for better solar forecasting to improve its integration into the energy supply. Fish-eye cameras are emerging in-situ meteorological sensors that have already demonstrated promising and interesting results for high temporal resolution and very short-term solar forecasting. However, current approaches to model the cloud cover dynamics from sky images still lack precision regarding the spatial configuration of clouds, their temporal dynamics and their physical interaction with solar radiation. The work described here aims at bringing innovative insights via a novel approach to irradiance forecasting using the Deep Learning framework, which constitutes an effective environment for a richer modelling of the cloud cover and its dynamics. The study shows that Convolutional Neural Networks (CNNs) are able to successfully estimate future irradiance from a sequence of past images of the sky. The corresponding 10-min forecast skill based on the Mean Square Error reaches 40% when evaluated on a set of 4000 unseen samples and shows an additional 10% performance improvement on the skill score, when past data of the same day are used to train the model. This outlines the need to incorporate historical data of the day in short term forecasting.