稀疏随机森林下的用电侧异常行为模式检测

发布时间：2018-01-29 20:12

本文关键词： 用电侧异常行为辨识随机森林随机权网络稀疏表示　出处：《电网技术》2017年06期 　论文类型：期刊论文

【摘要】：随着智能电网的不断推进与传感技术的高速发展,我国电网用电侧数据逐步呈现出高复杂度、高冗余度的幂指数增长态势。传统的用电行为模式检测技术已无法满足其分析处理需求。为此提出基于稀疏随机森林模型的用电侧异常行为模式检测方法。该方法首先利用时间窗函数与Bootstrap重采样,建立用电侧行为模式信息簇。其次,利用基于随机权网络的有监督学习得到随机森林模型。最后,对随机森林模型进行稀疏化,并依据异常积累量指标判定样本有无异常。仿真对比实验验证了提出的稀疏随机森林模型的精确性与高效性。此外,通过多种体量数据下的Hadoop分布式计算实验,验证了基于稀疏随机森林的用电侧行为模式检测方法对用电侧大数据的高效处理能力。
[Abstract]:With the continuous advance of smart grid and the rapid development of sensing technology, the power side data of China's power grid gradually presents a high complexity. The power exponent growth trend of high redundancy. The traditional detection technique of electric behavior pattern can not meet the demand of its analysis and processing. Therefore, a method of detecting abnormal behavior pattern of power side based on sparse stochastic forest model is proposed. The method firstly uses time window function and Bootstrap resampling. Secondly, the supervised learning based on stochastic weight network is used to obtain the stochastic forest model. Finally, the stochastic forest model is sparse. According to the index of abnormal accumulation, the sample is judged to be abnormal or not. The simulation results show that the proposed sparse stochastic forest model is accurate and efficient. The Hadoop distributed computing experiment based on a variety of volume data is carried out to verify the efficient processing ability of the method based on sparse stochastic forest to detect the behavior pattern of the power side of the power side big data.
【作者单位】：华北电力大学电气与电子工程学院;中国人民大学经济学院;
【基金】：国家863高技术基金项目(2015AA050203)~~
【分类号】：TM73;TM76
【正文快照】： 权网络的有监督学习得到随机森林模型。最后,对随机森林模型进行稀疏化,并依据异常积累量指标判定样本有无异常。仿真对比实验验证了提出的稀疏随机森林模型的精确性与高效性。此外,通过多种体量数据下的Hadoop分布式计算实验,验证了基于稀疏随机森林的用电侧行为模式检测方法

【相似文献】