Real-Time Fault Detection and Classification in Smart Grids Using Support Vector Machines

This research investigates real-time fault detection and classification in smart grids using five machine learning algorithms: Support Vector Machines (SVMs), k-Nearest Neighbors (k-NN' Decision Trees, Random Forest and Gradient Boosting. Extensive experiments were conducted to compare them performance-wise in terms of accuracy, computational efficiency and robustness. The findings showed that Gradient Boosting attained the highest accuracy of 95.2%, which surpassed other algorithms. Second, the RF was illustrated with competitive accuracy (94.6%) less overfitting providing a reliable option for use. Interpretable Decision Trees were challenged by issues surrounding overfitting. 92.5% accuracy was found in the case of SVMs; however, these turned out to be computationally complex procedures as well. K-NN reported competitive accuracy (89.3%), but was computationally expensive. Comparison to related works from all fields underscored the cross-domain relevance of fault detection in critical systems. The study provides useful observations on the choice of fault detection algorithms to be used in smart grids, which recommend a compromise based on accuracy performance and computational efficiency while also being adapted for specific grid features.