| Abstract [eng] |
This master’s thesis investigates the possibilities of identifying faults in electrical distribution networks using smart electricity meter (SMART) data. With increasing electrical grid loading, the growth of decentralized generation, and the large-scale deployment of smart metering infrastructure, early fault detection has become increasingly important for ensuring reliable electricity supply and reducing the impact of disturbances on consumers. The study is based on real smart meter measurement data comprising 92,861 measurements collected from 103 different meters during the period from 19 to 25 October 2025. The analysis focused on phase voltage and phase current measurements, which provide direct insight into grid operating conditions, load distribution, and potential abnormal behavior. A fault identification methodology was developed by combining phase imbalance analysis, the unsupervised machine learning algorithm Isolation Forest, and a reference-based similarity analysis approach. Three previously identified problematic smart meters were used as validation cases to form a reference profile representing a faulty operating condition. The remaining meters were then evaluated for similarity to this reference profile. The results demonstrated that the proposed methodology can effectively identify anomalies and potentially problematic meters even when only a limited set of parameters is available. While anomalies in the overall dataset were distributed among unrelated individual meters, the analysis of the known problematic cases confirmed the suitability of the approach for identifying real fault-related operating patterns. The findings show that smart meter data can be used not only for energy metering purposes but also for advanced power network diagnostics and early fault detection. |
