Entropy-Based and Deep Learning–Driven Detection of Stealthy Cyberattacks in Smart Distribution Networks

This study investigates and evaluates the performance and accuracy of advanced techniques for detecting concealed and stealthy cyber-attacks within the standard IEEE ۳۳-bus distribution network. Two representative attack scenarios, Replay Attack and Slowly-Varying/Low-Amplitude Stealth Attack, were simulated in the MATLAB environment to evaluate their impact on system operation. Simulation results indicate that these attacks induce gradual deviations and disturbances in the power flow and voltage profiles of buses No. ۶ and ۲۵. To identify such anomalies, two detection mechanisms were employed: a Shannon entropy–based approach and an autoencoder neural network architecture. The evaluation outcomes demonstrate that the entropy-based method provides high sensitivity and low latency in attack identification; however, it exhibits a higher false alarm rate (FAR) in response to minor fluctuations and normal network variations. Conversely, the Autoencoder architecture significantly reduces the false alarm rate and enhances detection accuracy for stealthy and progressive attacks. Ultimately, the hybrid integration of both methods yields a substantial improvement in the overall performance of the intrusion detection system (IDS) for smart distribution networks.

کلیدواژه ها:

Autoencoder ، Entropy shanon ، deep learning ، Stealthy Cyberattacks ، Smart Distribution Networks