摘要
为了有效量化负荷预测的不确定性,提出一种基于主成分分析PCA(principal component analysis)和高斯过程分位数回归GPQR(Gaussian process quantile regression)的负荷短期概率密度预测模型。首先,采用PCA对原始多维输入变量进行预处理,实现负荷原始数据的降维;其次,基于高斯过程回归模型,通过选择输入变量的主成分作为GPQR的输入,实现对预测区间完整的概率描述;最后,在不同分位数回归预测值的基础上,利用核密度估计输出任意时刻负荷的概率密度预测结果。所采用的PCA降低了模型训练复杂度,由于GPQR是一种贝叶斯方法,有效处理了电力负荷数据中的不确定性。通过与多种常规方法算例的测试对比,验证了所提模型的预测性能和有效性。
To effectively quantify the uncertainty in load forecasting,a novel short-term load probability density forecasting model based on principal component analysis(PCA)and Gaussian process quantile regression(GPQR)is proposed in this paper. Firstly,PCA is used to pre-process the original multi-dimensional input variables to realize dimensionality reduction of the original load data. Secondly,the principal components of input variables are selected as the input for GPQR based on a Gaussian process regression model,thereby realizing a complete probability description of the prediction interval. Finally,kernel density estimation is used to calculate the result of load probability density forecasting at any time based on different predicted values of quantile regression. The complexity in model training is reduced by PCA. In addition,since GPQR is a Bayesian method,it can effectively deal with the uncertainty in power load data.A numerical example is tested using different models,and the forecasting performance and effectiveness of the proposed model are verified from a comparison with other conventional methods.
作者
张建文
杨晨
冉懿
吕朋朋
缪平
ZHANG Jianwen;YANG Chen;RAN Yi;LYU Pengpeng;MIAO Ping(Electric Power Research Institute,State Grid Xinjiang Electric Power Co.,Ltd,Urumqi 830000,China;NARI Nanjing Control System Co.,Ltd,Nanjing 210000,China)
出处
《电力系统及其自动化学报》
CSCD
北大核心
2020年第5期24-29,54,共7页
Proceedings of the CSU-EPSA
基金
国家重点研发计划项目资助(2016YFB0901100)-“城区用户与电网供需友好互动系统”。
关键词
电网负荷
主成分分析
高斯过程分位数回归
概率密度预测
不确定性
power system load
principal component analysis
Gaussian process quantile regression
probability density forecasting
randomness
