Realisation of illegal electricity tracking and control system using machine learning algorithms

In this study, an innovative machine learning-based system is proposed for the detection of illegal loads in low voltage networks. The primary objective is to reduce electricity theft by identifying, classifying, and locating illegally connected loads, specifically in agricultural irrigation applications. To demonstrate the effectiveness of the proposed system, two pilot areas in the provinces of Elazığ and Malatya were modeled based on real field parameters of low voltage networks. A dataset was created by recording three-phase current data at the beginning of the feeder. Using this dataset, two different methods for detecting illegal connections were proposed. Features were extracted from single-phase and three-phase current data using Pattern Analysis, Discrete Wavelet Transform (DWT), and Fast Walsh-Hadamard Transform (FWHT). These extracted features were then used for training and testing machine learning algorithms. The performance of these algorithms was compared in classification and localization tasks, with ensemble methods demonstrating particular effectiveness. The proposed method's performance was further enhanced through grid search optimization, achieving high accuracy with real-world data. When features were extracted using the DWT method, the Optimizable Boosting Ensemble (OBE) algorithm was confirmed to produce the best results.