摘要
针对电力负荷大数据化越发突出,引入最小绝对值收敛及选择(Lasso)算法解决电力负荷大数据难题,对电力负荷及相关天气因素大数据进行高维数据特征提取,获得有用数据集。为避免输入空间严重自相关及网络维数较高,造成径向基函数(RBF)神经网络预测精度严重下降的不良影响,提出基于主元分析(PCA)改进的RBF神经网络电力负荷预测模型,消除多气象因素相关性,剔除冗余,提取天气因素特征量,将新天气特征量与历史负荷数据共同作为RBF网络的建模对象,既全面表征天气因素对电力负荷的影响,又简化预测模型,加快预测速率。经美国南部某地区实际电力负荷数据的预测分析,充分证明该方法的有效性及可靠性。
For big data feature of power load being more and more prominent, least absolute shrinkage and selection operator （Lasso） algorithm is introduced to solve the power load problem of big data by extracting features of high- dimensional data from power load and related weather factors. To avoid the adverse effect that the prediction accuracy of RBF neural network is affected severely, caused by that the autocorrelation of input space is too serious and the dimension of network is too high, the improved RBF neural network power load forecasting model is proposed based on principal component analysis. It eliminates the correlation among weather factors, excludes redundancy and extracts feature variables of multiple weather factors. The obtained weather characteristics are taken as the modeling objects of the RBF network together with the dates of historical load, which not only characterize fully the impact of weather factors on the power load, but also simplify the prediction model and accelerate the forecasting rate. Through the experiments of forecasting and analyzing to the actual power system load in a certain region of southern United States, it proved the validity and reliability of the method fully.
作者
张淑清
任爽
陈荣飞
钱磊
姜万录
李盼
ZHANG Shu-qing, REN Shuang, CHEN Rong-fei, QIAN Lei, JIANG Wan-lu, LI Pan(Institute of Electrical Engineering, Yanshan University, Qinhuangdao, Hebei 066004, Chin)
出处
《计量学报》
CSCD
北大核心
2018年第3期392-396,共5页
Acta Metrologica Sinica
基金
国家自然科学基金(51475405
61077071)
河北省自然科学基金(E2018203439
E2018203339)
河北省高等学校科技研究重点项目(ZD2014100)
关键词
计量学
短期负荷预测
电力负荷
大数据简约
主元分析
RBF神经网络
气象因子
metrology
short-term load forecasting
power system load
big data reduction
principal componentanalysis
RBF neural network
weather factors
