Short-Term Photovoltaic Power Prediction Based on Multi-Stage Temporal Feature Learning

Energy Engineering, Journal Year: 2025, Volume and Issue: 0(0), P. 1 - 10

Published: Jan. 1, 2025

Harnessing solar power is essential for addressing the dual challenges of global warming and depletion traditional energy sources.However, fluctuations intermittency photovoltaic (PV) pose its extensive incorporation into grids.Thus, enhancing precision PV prediction particularly important.Although existing studies have made progress in short-term prediction, issues persist, underutilization temporal features neglect correlations between satellite cloud images data.These factors hinder improvements performance.To overcome these challenges, this paper proposes a novel method based on multi-stage feature learning.First, improved LSTM SA-ConvLSTM are employed to extract spatial-temporal images, respectively.Subsequently, hybrid attention mechanism proposed identify interplay two modalities, capacity focus most relevant features.Finally, Transformer model applied further capture patterns long-term dependencies within multi-modal information.The also compares with various competitive methods.The experimental results demonstrate that outperforms methods terms accuracy reliability prediction.

Language: Английский

Spatio-Temporal Photovoltaic Power Prediction with Fourier Graph Neural Network

Electronics, Journal Year: 2024, Volume and Issue: 13(24), P. 4988 - 4988

Published: Dec. 18, 2024

The strong development of distributed energy sources has become one the most important measures for low-carbon worldwide. With a significant quantity photovoltaic (PV) power generation being integrated to grid, accurate and efficient prediction PV is an essential guarantee security stability electricity grid. Due shortage data from stations influence weather, it difficult obtain satisfactory performance prediction. In this regard, we present forecasting model based on Fourier graph neural network (FourierGNN). Firstly, hypervariable constructed by considering weather neighbouring plants as nodes, respectively. hypervariance then transformed in space capture spatio-temporal dependence among nodes via discrete transform. multilayer operator (FGO) can be further exploited information. Experiments carried out at six show that presented approach enables optimal obtained adequately exploiting

Language: Английский