分布式协同网络用电负荷分层加权概率预测方法  

Feature weighting based hierarchical probabilistic load forecasting in distributed collaborative network

作　　者：孙欣尧[1] 王雪[1] 吴江伟[1] 刘佑达[1] 

机构地区：[1]清华大学精密仪器系精密测试技术及仪器国家重点实验室,北京100084

出　　处：《仪器仪表学报》2014年第2期241-246,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61272428);教育部博士点基金(20120002110067)资助项目

摘　　要：先进计量体系(AMI)是智能电网中的分布式协同网络,其通过广泛布置的分布式测量计算节点对用电端的用户用电信息进行测量和协同分析。基于分布式协同网络测量得到的海量数据,针对短期用电负荷的概率预测问题提出一种分层特征加权概率预测方法。该方法采用核主分量分析提取用电负荷测量样本的非线性特征,根据提取的特征采用马氏距离判据对用电负荷数据进行特征加权,剔除权重低的不相关干扰数据;提出将经验模态分解与稀疏贝叶斯学习方法相结合的机器学习用电负荷概率预测方法,对用电负荷高频与低频分量进行分层概率分布预测。最后,将所提出的方法应用于某地区的短期用电负荷预测实验,实验结果表明该方法能够有效预测短期用电负荷的概率分布,预测精度高、可靠性好。Advanced metering infrastructure （AMI） is a type of distributed collaborative network, which measures and processes demand-side load information through extensively deployed sensing and computing nodes in smart grid. Based on the massive data collected by AMI, this paper explores a feature weighting hierarchical probabilistic predic- tion approach for short-term load probabilistic forecast. This method adopts kernel based principal component analysis to extract the non-linear characteristics of the load measurement samples;according to the extracted characteristics, Mahalanobis distance criterion is used to carry out the feature weighting of the load data and prune the uncorrelatcd interference data with low weight. A machine learning load probabilistic prediction method is proposed, which com- bines sparse Bayesian learning with empirical mode decomposition;and the hierarchical probabilistic distribution pre- diction for the high and low frequency components of the load is achieved. The proposed method was applied to the short-term load forecasting experiment in a certain district, and the experiment results illustrate that the proposed ap- proach exhibits better performance in comparison with the original SBL model, and has the advantages of high forecas- ting accuracy and high reliability.

关 键 词：分布式协同网络 用电负荷预测 特征加权 分层预测 稀疏贝叶斯学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]