Electricity theft detection model based on feature engineering and integrated classifier

Focused on tackling the issue because the electricity theft detection model has low applicability and the recognition rate is greatly affected by subjective factors like manual extraction features, a power-stealing user recognition model based on feature engineering and integrated classifier is proposed. The model uses Tsfresh and Whale Optimization Algorithm (WOA) as the main feature extraction and selection tool. Then we use an integrated classifier to identify the abnormal users. Firstly, we collected the electricity consumption data of 600 different users for 583 consecutive days. Secondly, the rough feature extraction is performed by Tsfresh and the variance-based method. Thirdly, the Whale Optimization Algorithm is used to finely select the features, getting the most suitable feature parameters. Finally, the integrated classifier is used to classify abnormal users. After evaluating the performance of different classifiers ensemble, an integrated classifier is composed of a Decision Tree (DT), Support Vector Machine (SVM), and Random Forest (RF). Meanwhile, in terms of feature extraction, we compared the results of three single classifiers which consist of the integrated classifier, also compared with other feature selection and classification methods. The experimental results show that the average accuracy of this method is 96.11%, which is higher than other methods.