Impact of Noise on Fault Classification in High-Voltage Transmission Lines Using LVQ Neural Networks

Abstract

Accurate fault detection and classification in high-voltage transmission lines are essential to ensure system reliability and operational safety. However, the presence of noise and transient disturbances often degrades the accuracy of conventional protection schemes. This study investigates the impact of Gaussian noise on fault classification performance using a neural network-based framework combined with Discrete Wavelet Transform (DWT) and Fast Fourier Transform (FFT) feature extraction. Four types of faults, single line to ground, line to line, double line to ground, and three phase to ground were simulated on a 150 kV transmission system using ATPDraw under various noise levels 40 dB. Linear Discriminant Analysis (LDA) and Learning Vector Quantization (LVQ3) were employed for feature reduction and classification, respectively. The proposed model achieved a test accuracy of 98.84% under free noise conditions and 96.80% under noisy conditions. This is outperforming traditional classifiers such as Support Vector Machine (SVM) and Decision Tree (DT). Results indicate that incorporating time-frequency domain features with noise-resilient neural architectures significantly enhances classification robustness and reliability. This research contributes a novel approach for noise-tolerant fault classification, offering practical potential for real-world implementation in intelligent protection systems and smart grid applications.