Official repository for MCloudNet, a multi-modal AI framework that enhances ultra-short-term photovoltaic (PV) power forecasting by modeling multi-layer cloud structures from satellite imagery.
MCloudNet integrates multi-layer satellite cloud images, ground-based meteorological data, and advanced deep learning techniques to improve PV power prediction, particularly in data-scarce rural micro-grids. The framework effectively captures cloud motion vectors, occlusion coefficients, and cloud layer interactions, leading to superior forecasting accuracy and interpretability.
✅ Multi-Layer Cloud Feature Extraction – Separates high-, middle-, and low-altitude cloud layers for enhanced cloud dynamics modeling.
✅ Multi-Modal Learning – Fuses satellite imagery and meteorological data to optimize forecasting accuracy.
✅ Robust Transferability – Achieves high performance even in new PV stations with limited historical data, leveraging satellite-based generalization.
✅ Energy & Economic Impact – Successfully deployed in 50+ photovoltaic stations across underdeveloped regions, reducing 60 million kWh of curtailment power and generating 24 million CNY in economic benefits.
If you find our work useful, please consider citing:
@article{MCloudNet2025,
title={MCloudNet: Multi-Layer Cloud Modeling for Ultra-Short-Term PV Forecasting},
author={Your Name et al.},
journal={IJCAI 2025},
year={2025}
}
git clone https://github.com/YourOrg/MCloudNet.git
cd MCloudNetpip install -r requirements.txtpython predict.py --input sample_dataMCloudNet is trained on real-world PV datasets, including:
🔹 Local Meteorological Data (LMD) – 15-min weather observations.
🔹 Numerical Weather Prediction (NWP) – High-resolution forecast data.
🔹 Satellite Cloud Imagery (Himawari-8) – Cloud-top temperature and dynamics.
🔹 Micro-Grid Energy Optimization – Enables reliable forecasting for off-grid PV systems in remote regions.
🔹 Sustainable Development – Supports clean energy expansion aligned with SDG 7, SDG 9, and SDG 13.
🔹 Scalability – Can be deployed in new PV sites without extensive retraining, making it highly adaptable for emerging markets.
