Abnormal Electricity Consumption Detection from Incomplete Records in Power System

Due to the limited channel bandwidth or interference signals in the advance metering infrastructures, there are usually some missing or human-revised data among the electricity consumption records of civilian customers. In order to make full use of this kind of records, machine learning techniques are introduced in this paper for electricity consumption sensitivity analysis regarding to the weather features. With the missing and revised records filtered out, each customer would have an individual regression model between weather conditions and the power demand. The importance of variables in the regression model is regarded as the sensitivity to various weather features. Then the abnormal consumption patterns are detected with a typical outlier identification algorithm based on different weather sensitivities among all the customers. The methods used in this paper show good results to identify the abnormal consumption patterns effectively regardless the quality of the original data.