Research on Abnormal Power Consumption Model Based on User Multidimensional Compound Features

Power loss is a serious problem for all power companies. To find effective means of abnormal electricity iidentification is a popular research field in recent years. This paper puts forward an anomalous electricity use model based on multi-dimensional compound features of electricity users. The support vector machine, local outlier factor, correlation measurement based on the similar user power load, and correlation change rate measurement based on the most relevant users—these four algorithms are adopted to extract four-dimensional compound features of anomalous electricity use from the perspective of global anomaly, local anomaly, regional space, and time sequence. Next, the logistic regression (LR) model is trained based on the compound features. After training, the LR model is adopted as the final anomalous electricity use identification model. Analysis of the practical power load data of users suggests that the LR model combine respective advantages of the four-dimensional compound features. Detection of anomalies using the LR model is an effective approach, which can reliably and accurately identify residents’ anomalous electricity use. From the accuracy rate, recall rate, precision rate and scores of F1, it can be seen that the LR model is significantly superior to SVM.