Turkish Journal of Electrical Engineering and Computer Sciences, vol.32, no.4, pp.605-622, 2024 (SCI-Expanded)
The decarbonisation of electricity generation requires real-time monitoring and control of grid components to efficiently and timely dispatch demand. This highly automated system, known as the Smart Grid, relies on smart or sensor-equipped distribution network components to optimise energy flow and minimise losses. However, energy theft, a major obstacle to efficient resource utilisation, poses a significant challenge to achieving this goal. This study proposes and evaluates a real-time telemetry and control system designed to mitigate energy theft in agricultural irrigation applications. The system increases energy efficiency by tracking the energy use in agricultural irrigation. The key challenge is to identify the source of illegal electricity consumption, classify it, and localise it. To address these difficulties, two distinct classification problems are addressed through the utilisation of machine learning methodologies. The initial classification task concerns the categorisation of loads that consume illegal electricity in agricultural irrigation. The subsequent classification problem pertains to the categorisation of feeder branches where such loads are activated. Therefore, a pilot distribution grid feeder has been simulated, and irrigation motors have been used as illegal loads which are activated at different points along the distribution feeder. The data collected from these simulations are used to create a data set where three-phase current data are collected from the transformer substation. The generated data set is employed to train machine learning models for the classification of illegal loads and feeder branches. The performance results of machine learning methods is obtained using the following metrics: accuracy, precision, recall, and F1-score. The results of the classification of loads stealing electricity in agricultural irrigation demonstrate that the bagged trees (BAT) algorithm achieves 99.64% in each criterion. In branch classification, the algorithm achieves the best results, with 97.64%, 97.40%, 96.22%, and 96.81%, respectively. Both classification performance results indicate that the proposed algorithm is effective for solving both classification problems. This research demonstrates the efficacy of ML-powered real-time monitoring and control in combating energy theft and promoting efficient resource utilisation within agricultural irrigation networks. It is a pioneering study in the field of determining and classifying illegal loads in agricultural irrigation. Further research will investigate the potential for expanding the system’s capabilities to include different load types and exploring alternative ML techniques for broader applicability within the context of low voltage distribution network monitoring.