Research On The Methods Of Data Anomaly Detection For Centralized Metering System

The centralized metering system has been applied to almost all low-voltage customers in our country,the aim of which is to improve the quality of metering and reduce the metering workload.However,abnormalities often occur in centralized metering data so that the data can't be used for electricity billing directly.So,how to discriminate the data abnormalities is the key to improving the accuracy of electricity billing and achieving the goal of reduce workers and increase revenues.Data anomaly determination includes defect determination and distortion determination.The former is aiming to identify the data defects(data missing / singular value / total consumption doesn't equal to the sum of peak and valley consumptions)caused by meter faults or poor communication quality,etc,while the latter aiming to discriminate the distortions of the metering data relative to the customers' real consumption not identified by the data defect analysis section or caused by electricity stealing activities.This thesis firstly presents a smoothness analysis method for centralized metering data defect analysis purposes,which can discriminate abnormal data uprushes from normal consumption increase.Suspected defective daily consumption series(DCS)are firstly searched out by smooth ratio analysis.The haar wavelet transformation is further applied to these suspected series and Lipschitz exponents are calculated for their singular points.Then true defect series are identified by searching singularities with Lipschitz exponents close to zero.Secondly,a consumption mode analysis based method is presented to overcome the shortcomings of the monthly consumption chain fluctuation ratio criterion now used for data distortion discrimination.A principal component analysis based method is presented for analyzing the customers' consumption mode stability.For the customers with stable consumption mode,the grey distance measure is further applied to evaluate their rational ranges of monthly electricity consumption,which are used as data distortion criteria.Thirdly,a multi-dimensional outlier detection method with tracing analysis functions is presented for electricity stealing suspect detection so as to furtherly ensure the accuracy of electricity billing.An example is provided towards one of the Electricity Supply Company located in Shanghai.Results show that the presented DCS smoothness analysis method is of greater defect discovery ability comparing with the current empirical criterion.The presented consumption mode analysis method can reduce the misjudgment in metering data distortion analysis caused by temperature variation or special consumption modes and reduce the amount of checked customers to reduce the checked workload effectively.The presented electricity stealing suspect identification method is of higher search rate with low misjudge rate,alleviating to a certain extent the concealment of electricity stealing in the environment of centralized metering.