Fault Detection and Classification in Power Transmission Lines Using Discrete Wavelet Transform-Based Swish Recurrent Neural Network

Transmission lines are an essential power grid component that is responsible for managing electricity from a production station to consumers and substations. The power network's transmission lines are primarily met with various faults. However, to maintain normal operation and minimize damage because of transmission line faults, it requires detecting and classifying faults. In this research, the Discrete Wavelet Transform-based Swish Recurrent Neural Network (DWT-SRNN) is proposed to detect and classify faults in power transmission lines using Deep Learning (DL). Initially, the data is acquired from the transmission line workspace by monitoring devices and sensors. Then, the Butterworth filtering is employed to increase the data quality. The DWT is performed to extract the features that transfer the time domain signal to the time-frequency signal during faulty and non-faulty conditions by decomposition. Finally, the SRNN is established to detect and classify faults in power transmission lines effectively. The existing techniques like Graph Convolutional Neural Network (GCNN), Machine Learning (ML), and Long Short-Term Memory (LSTM) are compared with DWT-SRNN. The proposed DWT-SRNN achieves a better accuracy of 99.89% compared to GCNN, ML, and LSTM respectively.

I. Introduction

The power transmission's fault detection and classification are the foundation of analysis and power accident treatment that is of primary significance to enhance the power grid stability [1]. Non-Technical Losses (NLTs) are regarded as energy which is flowed via electricity. A transmission line is a way of transmitting electrical power from producing power stations to distribution stations for domestic and commercial use. It is vulnerable to faults that are affected by events like tree falls because of lightning surges, strong winds, excessive loading conditions, equipment mechanical failures, and so on [2]. In modern power systems, the combination of inverter-based generators produces various difficulties in power system protection, control, stability, and planning. Asymmetrical fault current has been established as negative, positive, and zero-sequence elements in traditional rotating systems of machine power [3]. Therefore, wavelet transform has developed as an efficient signal-processing approach for the analysis of power transient disturbance systems. It results in voltage and current waveforms owning features like electromagnetic transients, time- varying frequencies, and trends [4] [5]. Furthermore, a High Impedance Fault (HIF) is one of the primary fault kinds in the system of distribution, and its fault faces a high threat to human safety [6]. Due to the development of the power industry, conventional manual approaches may no longer meet the requirements of modern grid power [7]. There are numerous opportunities for Artificial Intelligence (AI)-based Deep Learning (DL) and Machine Learning (ML) to maximize the performance of a grid ranging from greater accurate detection of load and renewable for real-time monitoring, planning, and protection [8]. The main contribution of this research is as follows: •

The data is gathered from the transmission line workspace by monitoring devices and sensors. These sensors capture real-time data regarding parameters like current, voltage, and temperature. Then, the Butterworth filtering is established to improve the data quality.

•

DWT is used to extract the features that can capture both time and frequency data which enables efficient analysis of transient signals connected with faults which provides multi-resolution features.

•

SRNN is utilized as a classification approach to detect and classify faults in power transmission lines effectively.