AdaBoost-SVM for Electrical Theft Detection and GRNN for Stealing Time Periods Identification
| Autor: | Liming Wang, Tianyu Hu, Rongli Wu |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Boosting (machine learning)
Training set Artificial neural network business.industry Computer science 020209 energy 02 engineering and technology Machine learning computer.software_genre Ensemble learning Support vector machine Statistical classification ComputingMethodologies_PATTERNRECOGNITION 0202 electrical engineering electronic engineering information engineering Artificial intelligence Electric power AdaBoost business Classifier (UML) computer |
| Zdroj: | IECON |
| Popis: | Electric power companies lose $96 billion every year because of non-technical losses. Electrical theft is the main part of non-technical losses. With the popularization of smart meters in the world, it is possible to detect suspect customers by data mining technology. In this paper, we propose a scheme that can not only find the anomalous users but also determine specific time periods of electric theft. In reality, abnormal customers account for a small part of all electric power users. Firstly, in terms of an imbalanced dataset, we propose an ensemble approach combining Adaptive Boosting algorithm (AdaBoost) and Support Vector Machine (SVM). The proposed scheme uses SVM as a weak classifier and changes the weight of the training data according to each training results until reach the threshold. Then the strong classifier is built after many iterations of the loop. We also conduct comparison experiments using four conventional machine learning approaches and two ensemble learning algorithms. The obtained results indicate the proposed method has better performance in the imbalanced dataset. Secondly, for abnormal consumers, we use General Regression Neural Network (GRNN) to estimate their electrical consumption and compare with the actual consumption to find the stealing electricity intervals. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|