一种电力感知数据的离群点检测方案

鉴于离群点引发的数据质量问题给电力应用造成的不良影响,对电力感知数据的特征进行了分析,并基于电力感知数据的时间特征和异常检测技术的易用性需求,提出一种电力感知数据的离群点检测方案。该方案由异常检测服务框架和离群点检测方法构成。异常检测服务框架借鉴Web服务的思想,基于大数据技术,能够支持电力感知数据的存储和计算,并且以服务的形式提供电力感知数据的异常检测能力。离群点检测方法是基于聚类算法和考虑时间属性的数据分段方法来检测电力感知数据中的离群点异常。通过实验验证了该方法的可行性和有效性,结果表明该方法能够有效识别具有时间相关性和连续性的电力感知数据中存在的离群点,且在数据规模增大时,具有良好的并行性和可扩展性。

An Electric Power Sensor Data Oriented Outlier Detection Solution

In view of the adverse effects of data quality problems caused by outliers on power applications,the characteristics of power sensor data are analyzed.Based on the temporal characteristics of power sensor data and the usability of anomaly detection technology,an electric power sensor data oriented outlier detection solution is proposed,which consists of an anomaly detection service framework and an outlier detection method.The anomaly detection service framework refers to the idea of Web service,and based on big data technology it can support the storage and calculation of power sensing data,and provide anomaly detection capability of power sensing data in the form of service.The outlier detection method is accomplished on the basis of clustering algorithm and a temporal characteristics related data segmentation method to detect outlier anomalies in power perception data.The feasibility and effectiveness of the proposed method are verified by experiment.The results show that this method can effectively identify outliers in power sensing data which are time-related and time-continuous,and has great parallelism and scalability when the data scale increases.