LSTM–GAN based cloud movement prediction in satellite images for PV forecast

Owing to the high uncertainty and variability of renewable energy, power system operators require an accurate forecast method. Considering that the cloud cover significantly affects the photovoltaic (PV) generation, critical factors for accurate PV forecast are the future shape and trajectory of clo...

Full description

Saved in:
Bibliographic Details
Published in:Journal of ambient intelligence and humanized computing 2023-09, Vol.14 (9), p.12373-12386
Main Authors: Son, Yongju, Zhang, Xuehan, Yoon, Yeunggurl, Cho, Jintae, Choi, Sungyun
Format: Article
Language:English
Subjects:
Accuracy
Algorithms
Artificial Intelligence
Cloud cover
Computational Intelligence
Deep learning
Engineering
Generative adversarial networks
Information services
Infrared imagery
Machine learning
Mathematical models
Meteorological data
Meteorological satellites
Neural networks
Original Research
Photovoltaic cells
Robotics and Automation
Satellite imagery
Support vector machines
Trends
User Interfaces and Human Computer Interaction
Variables
Weather forecasting
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Owing to the high uncertainty and variability of renewable energy, power system operators require an accurate forecast method. Considering that the cloud cover significantly affects the photovoltaic (PV) generation, critical factors for accurate PV forecast are the future shape and trajectory of clouds, which weather information services hardly provide. The paper proposes an innovative PV generation forecast method based on future cloud image prediction, for which a hybrid deep learning technique combining the generative adversarial network (GAN) and the long short-term memory (LSTM) model is used. The role of GAN is to generate cloud images from random latent vectors while LSTM learns patterns of time-series input images. To verify the effectiveness of the proposed methodology, the paper compares it with various hybrid PV forecast models in terms of prediction accuracy, using field data of satellite images and meteorological information. For testing the proposed method, a total of 30,507 infrared images shot by Communication, Ocean, and Meteorological Satellite 1 of the National Meteorological Satellite Center of Korea every 15 min were collected and utilized. In the end, it is concluded that the proposed LSTM–GAN model presents better prediction accuracy over CNN–ANN, CNN–LSTM, GRU–GAN, and BILSTM-GAN.
ISSN:1868-5137
1868-5145
DOI:10.1007/s12652-022-04333-7