Abstract
This study introduces a lightweight hybrid solar photovoltaic (PV) generation prediction model operating on 1-h intervals, utilizing remote sensing data to enhance power grid management. Multisource remote sensing data, including spatial features from infrared satellite images and temporal data from various hourly recorded datasets, capture spatiotemporal characteristics. The model defines and synthesizes regions of interest (ROI) and surrounding areas of ROI (ROIsurr) within satellite images to reduce computational load. Integration of image and numerical weather prediction (NWP) process modules ensures accurate prediction. Comparative analysis against five machine learning algorithms shows significant improvements, with up to a 33.7% decrease in mean absolute error (MAE) and a 19.51% decrease in root mean square error (RMSE). Additionally, the model consistently meets ASHRAE Guideline 14 standards and outperforms single-source data models. Experimentation highlights the effectiveness of smaller ROIs in enhancing predictive accuracy, demonstrating adaptability to climate variations. This lightweight multisource remote sensing-based hybrid model promises to guide smart grid operations and sustainable power grid systems, advancing remote sensing applications in renewable energy management.
Original language | English |
---|---|
Article number | 4704511 |
Pages (from-to) | 1-11 |
Number of pages | 11 |
Journal | IEEE Transactions on Geoscience and Remote Sensing |
Volume | 62 |
DOIs | |
State | Published - 2024 |
Keywords
- Convolutional neural network (CNN)-long short-term memory (LSTM)
- deep learning
- lightweight
- multisource data
- region of interest (ROI)
- satellite image
- solar photovoltaic (PV) generation forecasting
- spatiotemporal