Modern AI Technique Development For Fault Detection And Analysis On Distribution Network

The reliable and efficient operation of distribution networks is critical for modern power systems but increasing complexity and integration of distributed generation introduce novel uncertainties, leading to potential faults. This paper explores the application of Artificial Neural Networks (ANN) and Support Vector Machines (SVM) for fault detection and analysis within the IEEE 33 bus system. Both AI techniques were developed and trained using historical fault data, followed by a comparative analysis based on accuracy and confusion matrices. The ANN model demonstrated higher accuracy, achieving training and testing accuracies of 88.9517% and 88.3592%, respectively, but struggled to distinguish between single-line-to-ground (SLG) and double-line-to-ground (DLG) faults. In contrast, the SVM model achieved lower accuracies of 60.9123% (training) and 59.4459% (testing), with significant misclassifications across all fault types. The strengths and weaknesses of each approach are discussed, highlighting the ANN model's robustness and computational demands and the SVM model's efficiency in high-dimensional spaces but sensitivity to hyperparameter tuning. This study provides valuable insights into optimizing AI techniques for fault detection, aiming to enhance the reliability and efficiency of distribution networks.