摘要
为了提高当前电力计量系统中的电能量时间序列数据异常检测的准确率与效率,该文提出一种数据流驱动的电压三相不平衡异常检测方法。结合已有的面向数据流异常检测思路,实现对时间序列电能量数据的采样。采用基于循环神经网络的长短期记忆神经网络对数据建模预测,计算实际值与预测值之间的误差。采用指数加权平均的滑动窗口对误差进行平滑,实现了在过滤了三相电压数据正常波动的影响下判断某一时间段是否出现异常。试验结果证明,该方法在召回率和综合评价指标等有所提高,验证了该文方法对于电压三相不平衡的异常检测问题的可行性和有效性。
To improve the accuracy and efficiency of current electric energy time series data anomaly detection in power metering system,this paper proposes a data stream driven anomaly detection method for voltage three-phase unbalance.Based on existing data stream-oriented anomaly detection motivates,this paper implements sampling of time series electrical energy data.The existing long short-term memory(LSTM)neural network is used to model the data for prediction,and the error between the actual and predicted values is calculated.A sliding window with exponentially weighted average is used to smooth the error,and it is realized to determine whether an anomaly occurs at a certain time period with the influence of normal fluctuations in the three-phase voltage data filtered.The experimental results indicate that the method has improved in terms of recall rate and comprehensive evaluation index,which verifies the feasibility and effectiveness of the method for the anomaly detection problem of voltage three-phase unbalance.
作者
刘波
王大鹏
闫永昶
刘通宇
张园园
袁培森
Liu Bo;Wang Dapeng;Yan Yongchang;Liu Tongyu;Zhang Yuanyuan;Yuan Peisen(Supervision and Support Center of Power Supply Service of State Grid Mengdong Power,Tongliao 028000,China;State Grid Inner Mongolia East Power Company Ltd.,Hohhot 010010,China;College of Artificial Intelligence,Nanjing Agricultural University,Nanjing 210095,China)
出处
《南京理工大学学报》
CAS
CSCD
北大核心
2023年第2期245-253,共9页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金(61502236)
江苏省农业科技自主创新资金项目(SCX(21)3059)
上海市大数据管理系统工程研究中心开放基金(HYSY21022)。
关键词
电压三相不平衡
数据流
异常检测
循环神经网络
长短期记忆网络
指数加权移动平均
voltage three-phase unbalance
data stream
anomaly detection
recurrent neural network
long short-term memory
exponential weighted moving average