摘要
针对目前异常用电检测中存在的专变用户窃电率高、窃电模式难以察觉、使用窃电检测模型查找窃电用户过程中训练集不足的问题,提出了一种基于用电特征分析的无监督方式异常用电检测方法.该检测方法引入离群点查找算法,量化了海量数据中不同异常用电行为,将其提取为异常用电特征序列,并且根据专变用户不同计量方式和用电特点,构建了基于局部离群因子(local outlier factor,LOF)检测算法的用电不平衡特征序列,定义了考虑失压持续时间的电压异常特征序列和基于每日电流曲线聚类结果的电流异常特征序列,设计了异常用电评价流程,提出异常用电检测方法.通过实例验证了此检测方法能够完全甄别存在异常用电行为的用户,且检测方法对窃电时刻预测结果较为准确,其查准率和召回率的调和均值F1值为0.81.该方法能为企业提供异常行为发生的时间段,为窃电行为的及时调查提供有力的依据.
An unsupervised method for abnormal power consumption detection based on the characteristics analysis of power consumption is proposed in this paper according to the higher rate of power theft by special transformer users,the difficulty of the mode detection and the problem of insufficient training set in the process of using the detection model of electricity stealing to find the stealing-electricity users.The outlier finding algorithm is introduced in this detection method,and the different abnormal electricity usage behaviors are quantified in massive data.Meanwhile they are extracted into the feature series of abnormal electricity usage.Moreover,the unbalance feature series are constructed based on the detection algorithm of local outlier factor(LOF)according to the different measurement methods and the characteristics of power usage for the specialized users.After that,the feature sequence of voltage anomaly and current anomaly are defined considering the duration of voltage loss and the clustering results of daily current curve.Finally the evaluation process of anomaly power consumption is designed,and the detection method of anomaly power is proposed.The results show that the detection method can completely identify the users with the abnormal behavior of electricity consumption,and it is more accurate for the prediction results of the electricity stealing time.The mixed average value F1 of precision rate and recall rate is 0.81 and it can find the period of abnormal electricity consumption accurately.It also provides the stronger basis for the enterprises to the prompt investigation of electricity stealing behavior.
作者
黄悦华
郭思涵
鲍刚
程江洲
谌桥
王艺洁
HUANG Yuehua;GUO Sihan;BAO Gang;CHENG Jiangzhou;SHEN Qiao;WANG Yijie(College of Electrical Engineering&New Energy,China Three Gorges Univ.,Yichang 443002,China)
出处
《三峡大学学报（自然科学版）》
CAS
2021年第1期96-101,共6页
Journal of China Three Gorges University：Natural Sciences
基金
国家自然科学基金(61876097)。
关键词
用电特征
异常检测
用电信息采集系统
智能电网
离群点检测算法
electricity characteristics
anomaly detection
acquisition system of electricity information
smart grid
local outlier detecting algorithm
