To address the serious challenges to traditional power systems caused by high variability of solar production, especially the ramp events, this paper is aimed at predicting solar radiation accurately at a very-short-term scale, i.e. several minutes ahead. Quite different from traditional methods, our designed system conducts time series prediction based upon meteorological data and sky images' features with the Long Short-Term Memory (LSTM) network to learn the long-term dependency. Experiments have been conducted based on sky images' feature extracted by Convolutional Neural Network, meteorological data and solar geometric data with a large number of features collected over a seven-summer period. Experimental results show that the proposed forecasting system outperforms other methods in the literature, such as solar forecasting based upon optical flow tracking, Feedforward Neural Network (FNN) and Support Vector Regression (SVR), in terms of accuracy and robustness.
|Original language||English (US)|
|Title of host publication||2020 Asia Energy and Electrical Engineering Symposium, AEEES 2020|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|Number of pages||5|
|State||Published - May 2020|
|Event||2020 Asia Energy and Electrical Engineering Symposium, AEEES 2020 - Chengdu, China|
Duration: May 28 2020 → May 31 2020
|Name||2020 Asia Energy and Electrical Engineering Symposium, AEEES 2020|
|Conference||2020 Asia Energy and Electrical Engineering Symposium, AEEES 2020|
|Period||5/28/20 → 5/31/20|
Bibliographical noteFunding Information:
ACKNOWLEDGMENT This work was supported in part Fundamental Research Fund under JCYJ20170411102217994 and Guangdong grant No. 2017ZT07X152.
work was supported in part by Shenzhen Fundamental Research Fund under Grant No. JCYJ20170411102217994
© 2020 IEEE.
- Long Short-Term Memory (LSTM)
- photovoltaic power forecasting
- solar irradiance
- time series prediction