کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

A Multi-Branch Deep Learning Architecture with Hierarchical Fusion for Real-Time Short-Term Multi-Horizon PV Energy Forecasting

نویسندگان :

Sepehr Shakibi¹ Melika Vafaee Taj Khatooni² Mohsen Hamzeh³

1- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر 2- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر 3- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر

کلمات کلیدی :

Photovoltaic energy forecasting،deep learning،multi-branch architecture،hierarchical fusion،real-time prediction

چکیده :

Accurate short-term generation forecasting is crucial for efficient energy management and maintaining a stable power grid. This study introduces a multi-branch deep learning architecture with hierarchical fusion that effectively integrates the raw PV energy signal, wavelet-derived representations, environmental variables, and engineered temporal features. Evaluated on a real-world rooftop PV dataset, the model demonstrates superior prediction accuracy. Experimental results indicate that the proposed approach outperforms conventional LSTM, GRU, and CNN-LSTM models, achieving lower error rates and reduced variability in predictions. Feature ablation analysis reveals the key importance of energy and temporal inputs, while wavelet features and feature engineering contribute to improving forecast reliability. In addition, the inference process was deployed on a Raspberry Pi 5 to assess real-time performance, showing consistently fast and stable predictions across all samples. Overall, the proposed architecture offers a robust and practical solution for short-term energy forecasting under diverse real-world conditions.