Anomaly Detection for Overcurrent Flow in Smart Grid Systems Based on Smart Meter Data

Abstract

The modernisation of electricity distribution networks via smart grids presents new issues in monitoring and identifying abnormalities such as overcurrent flow which may arise from equipment malfunctions, unauthorised consumption or system inefficiencies. Conventional anomaly detection techniques rely on static thresholds are insufficient for contemporary smart grids due to their complexity and scale. This research proposes a machine learning approach for identifying overcurrent anomalies utilising smart meter data to overcome this gap. The study uses the Smart Meter Electricity Consumption Dataset from Kaggle, comprising power usage data at 30-minute intervals, environmental factors and pre-identified abnormalities. Data pre-processing, feature extraction and normalisation are executed in MATLAB succeeded by the assessment of several classifiers including Decision Trees, Random Forests and Neural Networks. Performance parameters such as accuracy, precision, recall and F1-score were used to evaluate the models. The Random Forest classifier achieves an AUC of 0.86 and an actual positive rate of 0.93 at a false positive rate of 0.08. The findings illustrate the model's effectiveness in detecting overcurrent incidents while reducing false positives. A statistical methodology employing moving averages and standard deviations establishes a criterion for comparison. The research highlights the potential of data-driven methods for enhancing grid dependability and advocates for the adoption of adaptive thresholds and hybrid models to drive future advancements. This study contributes to the overarching dialogue on smart grid security, offering practical recommendations for mitigating energy theft, enhancing maintenance efficiency and ensuring sustainable system functionality.