基于自适应特征提取与时间迁移建模的分布式光伏超短期功率预测

Accurate distributed photovoltaic power forecasting plays a crucial role in optimizing grid operations, enhancing economic benefits, and promoting the integration of new energy sources. However, existing methods for forecasting distributed photovoltaic power face several challenges: 1) Satellite cloud images can provide data support for distributed photovoltaics that lack specialized meteorological measurements, but the methods of cloud image features modeling tend to ignore important features; 2) Seasonal changes and variable climate conditions cause temporal distribution variations in photovoltaic output characteristics, leading to poor performance of trained forecasting models when there is a variation in data distribution, resulting in inadequate generalization capabilities. To address these issues, this paper proposes a regional ultra-short-term power forecasting method for distributed photovoltaic based on adaptive feature extraction and temporal transfer modeling. This approach integrates the spatial feature capture capability of Convolutional Neural Networks with the time-series processing mechanism of Transformer-based models to perform adaptive feature extraction of the correlation between multi-source remote sensing information and photovoltaic power. Subsequently, data distribution variations are quantified to divide the data into sequences with significant distribution differences. This allows the temporal transfer model to extract invariant generalized features in the temporal domain, thereby enhancing the model's generalization ability and forecasting performance. Finally, the effectiveness of the proposed method was validated using actual distributed photovoltaic power data.