基于相关性分析和长短期记忆网络分位数回归的短期公共楼宇负荷概率密度预测  

Short-term Public Building Load Probability Density Prediction Based on Correlation Analysis and Long-and Short-term Memory Network Quantile Regression

作　　者：杨秀 陈斌超 朱兰 方陈[2] YANG Xiu;CHEN Binchao;ZHU Lan;FANG Chen(College of Electric Engineering,Shanghai University of Electric Power,Yangpu District,Shanghai 200090,China;State Grid Shanghai Electric Power Research Institute,Hongkou District,Shanghai 200437,China)

机构地区：[1]上海电力大学电气工程学院,上海市杨浦区200090 [2]上海市电力公司电力科学研究院,上海市虹口区200437

出　　处：《电网技术》2019年第9期3061-3070,共10页Power System Technology

基　　金：国家自然科学基金项目(51807114)~~

摘　　要：公共楼宇是智能电网用电环节需求响应的重要组成部分,在强不确定性环境下,为了提高公共楼宇短期负荷预测的精度,并能更好反映楼宇负荷的不确定性。提出了一种集合多维尺度分析技术(multidimensional scaling,MDS),基于Copula函数相关性测度、长短期记忆网络分位数回归(quantile regression long short-term memory,QRLSTM)和核密度估计(kernel density estimation,KDE)的短期公共楼宇负荷概率密度预测的方法。首先采用MDS技术对楼宇群进行初步划分,再通过基于Copula函数的相关性测度方法定量计算影响因素(外界天气、人类活动)与目标楼宇负荷的相关程度;其次,运用QRLSTM回归模型预测未来不同分位数上的负荷值。最后,通过核密度估计得到未来任意时刻预测点的概率密度函数。实验结果表明,综合考虑强相关影响因素,并结合QRLSTM回归和KDE技术,能够更好地解决短期公共楼宇负荷概率密度预测问题。Public buildings are an important part of the demand response of smart grid electricity.In the environment of strong uncertainty,in order to improve the accuracy of short-term load forecasting of public buildings,and to better reflect the uncertainty of load,a new method of short-term public building load probability density prediction is proposed.The method is based on multidimensional scale analysis technology(MDS),Copula function correlation measurement,quantile regression long short-term memory(QRLSTM)and kernel density estimation(KDE).Firstly,the MDS technology is used to initially divide the buildings into groups,and then Copula function correlation measurement method is used to quantitatively calculate the correlation between influencing factors(weather,building occupancy and space available information)and building load changes.Secondly,the QRLSTM regression model is used to predict load values for different future quantiles.Finally,the probability density function of the predicted points at each moment is obtained by kernel density estimation.Experimental results show that considering the strong correlation influencing factors and combining QRLSTM regression and KDE technology,it can better solve the problem of short-term public building load probability density prediction.

关 键 词：楼宇负荷概率预测 强相关因素 多维尺度分析 COPULA函数 长短期记忆网络分位数回归 核密度估计 

分 类 号：TM721[电气工程—电力系统及自动化]