基于气象特征和改进Transformer的光伏功率短期预测  

Short-term Prediction of Photovoltaic Power based on Meteorological Features and Improved Transformer

作　　者：张建辉 滕婕 李秀慧 谭庄熙 ZHANG Jianhui;TENG Jie;LI Xiuhui;TAN Zhuangxi(Economic and Technological Research Institute of State Grid Gansu Electric Power Co.,Ltd.,Lanzhou,China,Post Code:730030;School of Information and Electrical Engineering,Hunan University of Science and Technology,Xiangtan,China,Post Code:411201)

机构地区：[1]国网甘肃省电力公司经济技术研究院,甘肃兰州730030 [2]湖南科技大学信息与电气工程学院,湖南湘潭411201

出　　处：《热能动力工程》2024年第8期146-154,共9页Journal of Engineering for Thermal Energy and Power

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

摘　　要：光伏出力易受气象因素影响,从而呈现间歇性和随机性。而准确可靠地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。本文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;然后将提取的特征和光伏出力数据输入BOA-iTransformer模型,再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;随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。实验结果表明,本文模型比iTransformer、Transformer和LSTM模型预测精度分别提高了3.54%,7.24%和14.2%。Photovoltaic(PV)output is susceptible to meteorological factors,thus showing intermittency and randomness.Accurate and reliable prediction of PV power can not only alleviates the impact of high percentage of PV grid-connectedness on the power grid,but also provides data reference for grid scheduling decision makers.In this paper,we proposed a short-term prediction method of PV power based on meteorological features and improved Transformer.Firstly,incremental features,statistical features and time-varying features were extracted for PV-related meteorological factors;then,the extracted features and PV output data were input into the BOA-iTransformer model,and each variable was embedded independently,which was convenient for the model to capture the key meteorological features and the correlation of multivariate data;subsequently,Bayesian optimal tuning was used for feature selection to obtain the optimal feature combinations,which was used to build the BOA-iTransformer PV prediction model;finally,the actual data of photovoltaic power stations in a region of China were used for comparative experi ments.The experimental results show that the prediction accuracy of this model can be improved by 3.54%,7.24%and 14.2%compared with iTransformer,Transformer and LSTM models,respectively.

关 键 词：光伏预测 深度学习 贝叶斯优化 特征构造 

分 类 号：TM615[电气工程—电力系统及自动化]