基于数据集蒸馏的光伏发电功率超短期预测  

Ultra-short-term Prediction of Photovoltaic Power Based on Dataset Distillation

作　　者：郑珂 王丽婕[1] 郝颖 王勃 ZHENG Ke;WANG Lijie;HAO Ying;WANG Bo(School of Automation,Beijing Information Science and Technology University,Haidian District,Beijing 100192,China;China Electric Power Research Institute,Haidian District,Beijing 100192,China)

机构地区：[1]北京信息科技大学自动化学院,北京市海淀区100192 [2]中国电力科学研究院有限公司,北京市海淀区100192

出　　处：《中国电机工程学报》2024年第13期5196-5207,I0015,共13页Proceedings of the CSEE

基　　金：北京信息科技大学“勤信人才”培育计划(QXTCP C202107)。

摘　　要：云是影响太阳直接辐射变化的主要因素,由于各类云的透光率不同,导致到达光伏电站的太阳辐射会随之产生波动。为解决各类云遮挡下的光伏发电功率波动大、预测模型个数多的问题,提出一种基于卫星云图和数据集蒸馏的光伏发电功率超短期预测模型。首先,基于待测场站上方的历史云图,采用Farneback光流法预测出云图;然后,根据卫星云分类标签数据建立各类云的样本库,利用数据集蒸馏算法训练样本库得到云类判别图,将预测云图与云类判别图匹配计算,获得云类聚合匹配特征;最后,利用上述特征、云量特征以及数值天气预报数据建立长短期记忆网络模型,对光伏发电功率进行超短期预测。利用某光伏电站数据进行验证,结果显示,该文所提模型能准确描述云层的各项特征,有效提升光伏功率预测精度。Cloud is the main factor affecting the change of direct solar radiation.Due to the different transmittance of various clouds,the solar radiation of photovoltaic power station will fluctuate accordingly.In order to solve the problems of large fluctuation and large number of prediction models of photovoltaic power generation under various clouds,an ultra-short-term prediction model of photovoltaic power generation based on satellite cloud image and data set distillation is proposed.First,based on the historical cloud image above the station to be measured,the Farneback optical flow method is used to predict the cloud image.Then,the sample library of all kinds of clouds is established according to the satellite cloud classification label data,and the cloud class discriminant map is obtained by training sample library of the data set with distillation algorithm.The predicted cloud image is matched with the cloud class discriminant map to obtain the cloud class aggregation matching feature.Finally,the long short-term memory network model is established by using the above features,cloud cover feature and numerical weather forecast data to predict the ultra-short-term photovoltaic power generation.The results show that the proposed model can accurately describe the characteristics of clouds and effectively improve the prediction accuracy of photovoltaic power.

关 键 词：数据集蒸馏 卫星云图 云分类 光流法 超短期光伏功率预测 

分 类 号：TM615[电气工程—电力系统及自动化]