光伏出力易受气象因素影响，从而呈现间歇性和随机性。可靠准确的预测光伏出力不仅可以缓解高比例光伏并网对电网的冲击，还可以为电网的调度决策人员提供数据参考。因此，提出一种基于气象特征和改进Transformer的光伏功率短期预测方法。首先针对光伏相关的气象因素提取增量特征、统计特征和时变特征。然后和光伏出力数据输入BOA-iTransformer模型，将每个变量独立嵌入，便于模型捕捉关键气象特征和多元数据的关联性；然后采用贝叶斯优化调参进行特征选择，得到最优特征组合，以此建立BOA-iTransformer光伏预测模型。最后采用中国某地区实际光伏发电站数据进行对比实验，算例结果表明所提方法相较对比方法iTransformer、Transformer和LSTM模型预测精度可分别提高3.54%、7.24%和14.2%。

Photovoltaic (PV) power is susceptible to meteorological factors, thus showing intermittency and randomness. Reliable and accurate prediction of PV power can not only mitigate the impact of high percentage of PV grid-connectedness on the power grid, but also provide data reference for the scheduling decision makers of the power grid. Therefore, a short-term prediction method of PV power based on meteorological features and improved Transformer is proposed. Incremental features, statistical features and time-varying features are first extracted for PV-related meteorological factors. Then, the PV power data are input into the BOA-iTransformer model, and each variable is embedded independently, which is convenient for the model to capture the key meteorological features and the correlation of the multivariate data; then, Bayesian optimal parameter tuning is used for feature selection, and the optimal feature combinations are obtained, so as to build the BOA-iTransformer PV prediction model. Finally, the actual PV power plant data in a region of China is used for comparison experiments, The example results show that the proposed method can improve the prediction accuracy by 3.54%, 7.24% and 14.2% compared to the comparison methods iTransformer, Transformer and LSTM models respectively.

国网总部科技项目(52272810005)；湖南省自然科学基金（2022JJ40150）