摘要
由于现有月度用电量预测所选影响因素较少,无法较为全面地反映与用电量强关联的因素,同时针对高维数据变量筛选和高精度预测等突出难题,文中提出了一种弹性网络用电量预测模型。为了考虑更为全面的影响因素,建立了用电量、气象、经济、交通4类,共340个变量的数据集。首先对8年96个点的高维变量数据进行弹性网络因子筛选,然后使用Granger因果关系分析找出了用电量数据与其它数据的关联关系,对一年范围内的全社会月度用电量使用弹性网络进行预测,预测结果的平均绝对百分误差为3.07%。为验证该模型的有效性,对比向量自回归(VAR)模型,反向传播(BP)模型和最小绝对值收缩和选择算子(Lasso)预测的效果,验证了文中所提方法预测精度较高。
Since the existing monthly electricity consumption forecast has fewer influencing factors,and it is unable to comprehensively reflect the factors associated with strong electricity consumption.An elastic network electricity consumption forecasting model for high-dimensional data variable screening and high-precision prediction is proposed.The volume prediction model analyzes the monthly data of 340 variables and 96 time points for electricity consumption,economy,transportation,and meteorology.By using elastic network to screening for high-dimensional variables,and Granger causality analysis to find out the dependence of electricity consumption data and other data,the monthly electricity consumption of the whole society in a year is predicted.And the mean absolute percentage error of the prediction results is 3.07%.Compared with the VAR model,BP model and Lasso,the feasibility and effectiveness of the method are verified.
作者
胡春凤
田世明
苏航
HU Chunfeng;TIAN Shiming;SU Hang(China Electric Power Research Institute,Beijing 100192)
出处
《电力工程技术》
2020年第3期166-172,共7页
Electric Power Engineering Technology
基金
国家电网有限公司总部科技项目“支持电力大数据分析的核心算法改进及其实用化技术”(520940180016)。
关键词
弹性网络
最小绝对值收缩和选择算子
GRANGER因果关系
因子筛选
用电量预测
elastic network
least absolute shrinkage and selection operator(Lasso)
Granger causality
factor screening
electricity consumption forecasting