Single-phase fault diagnosis in distribution networks via cyber-secure hybrid models of artificial intelligence

Ensuring stability and cyber-security in modern distribution networks is critical for the reliable operation of smart grids. This study proposes a hybrid artificial intelligence model that integrates an Autoencoder (AE) for feature compression and denoising, a Convolutional Neural Network (CNN) for spatial feature extraction, and a Bidirectional Long Short-Term Memory (BiLSTM) network for temporal dependency learning, collectively forming the AE-CBiLSTM model. The model is designed for accurate diagnosis of single-phase-to-ground (SPG) short-circuit faults in distribution systems. Simulations are conducted on the IEEE 13-bus test feeder using EMTP, with data processing and model implementation performed in MATLAB. The proposed model is evaluated under two operational scenarios: normal conditions and cyber-attacks in the form of False Data Injection Attacks (FDIA). Comparative results against various benchmark models demonstrate the superior accuracy and F1-score of the AE-CBiLSTM model, especially under adversarial data conditions. The model shows strong resilience and reliability, maintaining high performance even when measurement data is compromised. These findings highlight the potential of AE-CBiLSTM as a secure and effective solution for real-time fault detection in distribution networks.