Enhanced Fault Classification, Detection, and Location Estimation in the IEEE 14-Bus Smart Grid Using a Hybrid CNN-LSTM Algorithm with Adaptive Learning Rate

Abstract

Smart grids play a crucial role in ensuring reliable and sustainable energy delivery. However, the increasing complexity of transmission and distribution networks, driven by rising energy demands, presents significant challenges in fault detection, classification, and localization. Accurate and timely identification of faults both symmetrical and asymmetrical is essential to mitigate the long-time power outages and maintain grid stability when isolating the faulty part. This study proposes an advanced fault diagnosis approach using a hybrid convolutional neural network (CNN) and long short-term memory (LSTM) model. Currents and voltages data were collected from an Institute of Electrical and Electronics Engineers IEEE 14-bus test system, with various fault scenarios simulated in MATLAB Simulink. Without additional preprocessing, the raw of three-phase voltages and currents signals were directly input into a cascade-parallel reconstructed CNN utilizing varying filter lengths to capture diverse signal features. The extracted multi-scale features were then processed by the LSTM network to learn time-related dependencies effectively. The proposed hybrid CNN-LSTM methodology outperforms conventional approaches in fault analysis, achieving high accuracy across various fault diagnosis tasks. The study introduces three distinct classifiers: fault class type Identification, achieving 99.98% accuracy, line faulty section identification, with an accuracy of 96%, and fault location estimation, with 99.98% accuracy. The model's effectiveness is further validated using key performance metrics, including accuracy, precision, recall, and F1-score, demonstrating its robustness and reliability for real-time smart grid fault analysis.