Deep learning and wavelet packet transform for fault diagnosis in double circuit transmission lines
Ziad M. Ali, Ehab M. Esmail
Abstract
Fault diagnosis in double-circuit transmission lines (DCTLs) involves fault detection, section identification, and accurate location, critical components in ensuring robust protection schemes. This paper proposes an advanced directional protection framework that integrates wavelet packet transform (WPT) with deep learning (DL) models, utilizing double-ended measurements of three-phase currents and voltages. The system is modeled using a distributed parameter line representation that includes shunt capacitance. The WPT technique extracts approximation coefficients from current and voltage signals, which serve as inputs to deep learning models, particularly using the mother wavelet packet db10 for optimal decomposition. The proposed method comprises two main stages: (i) detection and identification of the faulted section and direction, and (ii) estimation of the fault location from the relaying point. The approach is evaluated using multiple deep learning architectures, including convolutional neural networks (CNNs) and recurrent neural networks (RNNs), and implemented in MATLAB. Simulation results across diverse fault scenarios, varying in location, resistance, and inception angle, demonstrate high accuracy and robustness. Compared with other hybrid approaches integrating WPT with artificial neural networks (ANNs) and adaptive-network-based fuzzy inference systems (ANFIS), the proposed method achieves superior precision, with an average error of only 0.03%. Notably, the technique offers primary protection for most line sections and backup coverage for adjacent areas.