考虑时空相关性的区域分布式光伏功率超短期预测

打开文本图片集

doi： 10.3969/j .issn.1673-5862.2025.03.010

Ultra-short-term forecasting of regional distributed photovoltaic power based on spatiotemporal correlation

PENG Yiming1，LI Dongxu1 ，LUO Yuanrong1 ，ZHOU Changcheng1 （1. Guangzhou Power Supply Bureau，Guangdong Power Grid Co.，Ltd.，CSG，Guangzhou 510oo0，China; 2.Southern Power Grid Digital Grid Research Institute Co.，Ltd.，Guangzhou 5loooo，China）

Abstract： In response to the issue that existing research predominantly focuses on enhancing predictive performance through the introduction of high-performance deep learning models，often lacking a principled explanation for the rationale behind model incorporation，this paper first analyzes the spatiotemporal correlations of power generation among regional photovoltaic （PV） power plants. It elucidates that due to the influence of cloud dynamics， the fluctuations in photovoltaic power exhibit a certain degree of spatial continuity. Building upon this foundation，a spatiotemporal graph attention network is introduced to effectively capture this complex spatiotemporal correlation. Specifically， the approach involves：1） utilizing a graph attention network to aggregate spatial information of regional PV power generation； 2） incorporating a onedimensional convolutional neural network to enhance the models feature extraction capabilities；and 3） employing gated recurrent units to aggregate temporal information，ultimately yielding the predicted power output. Empirical results demonstrate that the proposed model achieves higher accuracy compared to traditional CNN and LSTM models. Furthermore，by extracting the attention weights of the model，it is shown that the model is capable of capturing the spatiotemporal correlations between regional power plants to a certain extent，thereby elucidating the necessity of model incorporation and the reasons for the improvement in predictive accuracy.

Key Words： photovoltaic power generation prediction； spatiotemporal correlation； graph attention network；gated recurrent unit；interpretability

传统能源的逐渐枯竭与人们对电力需求的增长形成了尖锐矛盾。（剩余7510字）