【摘要】 整县光伏政策促使小容量屋顶光伏急剧增长，实现屋顶分布式光伏超短期发电功率的准确预测是分析海量细粒户用光伏电站对电力系统影响的前提。然而，屋顶分布式光伏在原有波动性的基础上存在小容量、分散式、离线式经营的特点，同时缺乏准确的气象数据，使得光伏功率预测异常复杂。为此，文章在有限数据下纵向地从光伏系统历史功率数据中搜索相似样本，横向地收集相邻分布式光伏发电用户功率数据，实现双向数据扩充，在一定程度上克服了光伏发电预测对于一些关键输入特征的依赖；在此基础上借助LSTNet(Long-and Short-term Time-series Network)神经网络的短期局部特征捕捉、长期时序信息强化、周期线性成分提取功能实现光伏功率预测。实验结果表明，在缺乏重要辐照数据的情况下，所提模型仍具有较好的预测精度。更多还原

【Abstract】 China’s "Whole County PV" programme has been dramatically expanding the use of solar power in rural areas, by building on government, comnmercial, industrial and residential rooftops. However, a large number of dispersed residential PV will have an impact on the power system, and accurately predicting the short-term power generation of residential PV is a prerequisite for addressing the impact. However, in addition to its original volatility, residential rooftop PV also has the characteristics of small capacity, decentralized and offline operation,together with the lack of accurate meteorological data, making PV power prediction exceptionally complex. Therefore,under the limited data, this paper longitudinally detects similar samples from the previous power data of the residential PV to be predicted,and horizontally collects similar samples from the power data of neighboring residential PV, ultimately jointly realizing two-dimensional data expansion, which overcomes the dependence of PV power generation prediction on some key input features to a certain extent. And then a residential PV generation prediction method is proposed based on LSTNet neural network, which has the functions of short-term local features capture, long-term time series information reinforcement, and cyclical linear component extraction.更多还原