基于双层XGBoost和数据增强的空间负荷预测方法  

Spatial load forecasting method based on double-layer XGBoost and data enhancement

作　　者：黄冬梅 张宁宁 胡安铎 胡伟 肖勇 陈岸青 HUANG Dongmei;ZHANG Ningning;HU Anduo;HU Wei;XIAO Yong;CHEN Anqing(College of Electronic and Information Engineering,Shanghai University of Electric Power,Shanghai 201306,China;College of Economics and Management,Shanghai University of Electric Power,Shanghai 201306,China;College of Continuing Education,Shanghai University of Electric Power,Shanghai 200090,China;State Grid Info-Telecom Great Power Science and Technology Co.,Ltd.,Fuzhou 350001,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海201306 [2]上海电力大学经济与管理学院,上海201306 [3]上海电力大学继续教育学院,上海200090 [4]国网信通亿力科技有限责任公司,福建福州350001

出　　处：《电力工程技术》2023年第1期201-208,共8页Electric Power Engineering Technology

基　　金：上海市科委地方院校能力建设项目(20020500700)资助。

摘　　要：为了解决空间负荷预测面临的特征变量众多和数据匮乏问题,文中提出一种基于双层极端梯度提升(extreme gradient boosting,XGBoost)和数据增强的空间负荷预测方法。该方法首先将待预测区域按馈线供电范围划分为若干子区域;其次构建基于双层XGBoost的特征选择模型,第一层XGBoost对特征进行评分及排序,将组合特征和负荷输入第二层XGBoost并进行子区域负荷预测,根据子区域负荷预测结果选择每个子区域的最佳特征变量;然后利用生成对抗网络(generative adversarial network,GAN)增强每个子区域的训练集样本,并通过极限学习机(extreme learning machine,ELM)实现子区域预测;最后将每个子区域的预测值相加得到待预测区域的预测值。以上海市局部区域为例,对文中方法进行仿真实验和对比分析。结果表明,文中方法可同时解决特征变量选择和数据匮乏问题,具有更高的预测精度。Spatial load forecasting faces the problems of multiple characteristic factors and data shortage.A spatial load forecasting method based on double-layer extreme gradient boosting(XGBoost)and data enhancement is proposed.Firstly,the area to be predicted is divided into several sub regions according to the supply range of feeder power.Secondly,a feature selection model based on double-layer XGBoost is constructed.The first layer XGBoost scores and sorts the features.The combined features are loaded into the second layer XGBoost for sub regional load forecasting.The best feature variables of each sub region are selected according to the load forecasting results.Then,the training set samples of each sub region are enhanced by the generative adversarial network(GAN),and the load of sub regions is forecasted through the extreme learning machine(ELM).Finally,the predicted values of sub regions are added to obtain the load of the region to be predicted.Taking local areas of Shanghai as an example,the simulated experiment and comparative analysis are carried out.The results show that the proposed method can solve the problems of characteristic variable selection and data shortage at the same time,and has high prediction accuracy.

关 键 词：空间负荷预测 极端梯度提升(XGBoost) 特征选择 生成对抗网络(GAN) 数据增强 极限学习机(ELM) 

分 类 号：TM715[电气工程—电力系统及自动化]