Balanced Attention-BiGRU Based Fault Detection and Classification in Electric Power System

Abstract

The modern era's expanding demand for electrical power, along with limited production and transmission capacity, has underlined the vital need for strong and rapid fault detection mechanisms inside electrical power networks. These intricate and dynamic systems, which rely primarily on transmission lines, are prone to disruptions and breakdowns induced by unanticipated factors. The work proposes a novel approach to addressing long-standing concerns in Artificial Intelligence (AI)-driven research, such as data imbalance, model complexity, computing cost, and over fitting issues. To solve these issues and improve the efficacy of AI-driven research, this study proposes an attention-based Bidirectional Gated Recurrent Unit (Bi-GRU) model designed specifically aimed for balanced datasets. The suggested approach initiates by preparing the dataset through normalization and one-hot encoding. Subsequently, the pre-processed dataset undergoes balancing using Synthetic Minority Oversampling Technique (SMOTE) before being inputted into the Attention-based Bi-GRU model. This model is designed to accommodate a fixed-length sequence of voltage and current values at its input layer. Finally, the batch normalization and dense layer with SoftMax were utilized to classify the types of defects in transmission lines. As a result, the suggested method classifies fault types with a high accuracy of 85.40% when compared to existing models.