模式聚类与周期分解的能耗监测及异常检测方法  

Research on energy consumption monitoring and anomaly detection method based on pattern clustering and periodic decomposition

作　　者：金静 方园 费洋 魏源 Jin Jing;Fang Yuan;Fei Yang;Wei Yuan(Xi′an Rail Transit Group Company Limited,Xi’an 710018,China;NARI Rail Transit Technology Co.,Ltd.,Nanjing 210061,China)

机构地区：[1]西安市轨道交通集团有限公司,西安710018 [2]南瑞轨道交通技术有限公司,南京210061

出　　处：《电测与仪表》2023年第11期151-157,共7页Electrical Measurement & Instrumentation

基　　金：国家重点研发计划项目(2022YFB2404000);江苏省产业前瞻与关键核心技术资助项目(BE2020081)。

摘　　要：时间序列异常检测对于能耗监测和管理至关重要。为了解决周期分解算法在时间序列中易受模式不一致影响的问题,文章提出模式聚类和周期分解相结合的异常检测方法。该方法使用密度峰值聚类算法寻找聚类中心点,并通过循环距离度量方法解决子序列相位偏移问题;为了屏蔽不同模式数据间的影响,使用季节分解算法S-H-ESD(seasonal hybrid extreme studentized deviate)在具有相同周期模式的聚类簇中寻找异常点,并评估和筛选检测结果的异常程度。实验结果表明,方法得到了较为理想的效果,适用于复杂周期时间序列的异常检测工作。Time series anomaly detection is crucial for energy consumption monitoring and management.In order to solve the problem that the periodic decomposition algorithm is easily affected by the inconsistency of patterns in time series,this paper proposes an anomaly detection method combining pattern clustering and periodic decomposition,which uses a density peak clustering algorithm to find the cluster centers and uses the cyclic distance measurement to solve the sub-sequence phase shift.To shield the influence between different patterns,the seasonal decomposition algorithm S-H-ESD(seasonal hybrid extreme studentized deviate)is used to find anomalies in clusters with the same periodic pattern,and the anomaly degrees of the detection results are evaluated and filtered.The experimental results show that the proposed method is effective and suitable for anomaly detection of complex periodic time series.

关 键 词：时间序列 能耗异常 模式聚类 周期分解 

分 类 号：TM933[电气工程—电力电子与电力传动]