考虑信息动态表达的异常用电模式识别云边协同方法

Cloud-edge collaboration method for abnormal power consumption pattern recognition considering dynamic expression of information

异常用电识别是用电稽查、计量装置运行状态辨识的重要内容,对维护电网的安全运行和保障正常用户权益有重要的意义。已有方法为了识别用户的多元用电模式,在保证识别准确性的基础上容易造成计算过于复杂的问题,而考虑效率的简单计算方法又难以准确度量不同用电模式的相似性,因此难以兼顾计算效率与准确性;此外,将用电数据上传至云端集中计算会占用大量的网络带宽和计算资源,进一步限制了异常辨识的应用。为此,提出了一种考虑信息动态表达的异常用电模式识别云边协同方法。根据边缘端和云端的计算资源合理分配协作任务,实现了异常用电的云边协同识别。针对边缘服务器算力有限的问题,对用电数据进行动态压缩重表达,在缩减数据量的同时保证数据信息的准确性。云端在收到压缩数据后以分段加权动态时间规整距离作为压缩数据相似性度量的依据,基于自适应参数选择的密度聚类算法识别异常用电。基于实际数据集验证了所提方法的有效性。

Abnormal power consumption recognition is an important part of the power consumption check and the operation status identification of metering devices,and is of great significance to maintain the safe operation of power grid and protect the rights and interests of normal users.In order to recognize multiple power consumption patterns of users,existing methods tend to cause complicated calculation on the basis of ensuring recognition accuracy.While,simple calculation methods considering efficiency are difficult to accurately measure the similarity of different power consumption patterns,so it is difficult to give consideration to calculation efficiency and accuracy.In addition,uploading power consumption data to the cloud for centralized computing consumes a large amount of network bandwidth and computing resources,which further limits the application of anomaly recognition.Therefore,a cloud-edge collaboration method for abnormal power consumption pattern recognition considering the dynamic expression of information is proposed.According to the computing resources of the edge terminal and the cloud,the cooperative tasks are reasonably allocated to realize the cloud-edge collaboration recognition of abnormal power consumption.Aiming at the problem of limited computing power of edge servers,the dynamic compression and re-expression of power consumption data are carried out to reduce the data amount and ensure the accuracy of data information.After receiving the compressed data,the cloud takes the segmented weighted dynamic time warping distance as the basis for the similarity measurement of compressed data,and identifies abnormal power consumption based on the density clustering algorithm with adaptive parameter selection.The effectiveness of the proposed method is verified based on the actual data set.