Unsupervised Learning With Hybrid Models for Detecting Electricity Theft in Smart Grids

Abstract

In the energy sector, electricity theft presents serious financial and security risks. By fusing supervised learning models (Random Forest) with unsupervised learning algorithms (Isolation Forest, One-Class Support Vector Machine (SVM), Local Outlier Factor (LOF), and Density-Based Spatial Clustering of Applications with Noise(DBSCAN)), this study presents a unique hybrid technique for identifying power theft. The models are developed and tested by the study using data from the State Grid Corporation of China (SGCC). Data on power use is examined by unsupervised algorithms to find abnormalities, which are then further examined by the Random Forest classifier for increased precision. The hybrid models work well in identifying anomalous consumption patterns that point to theft without requiring large amounts of labeled data. To improve grid sustainability and lower non-technical losses, this study offers power providers a scalable and effective option. The study advances the discipline by providing an original model detection and demonstrating its potential application in practical scenarios.

Keywords

• Electricity theft detection
• smart grids
• unsupervised learning
• hybrid models
• anomaly detection
• supervised learning