A New Deep Learning-Based Strategy for Launching Timely DoS Attacks in PMU-Based Cyber-Physical Power Systems

Abstract

Malicious attacks in the cyber-physical power systems (CPPS) can eventually result in cascading failure and even widespread blackout, if not rectified in a timely manner. The probability of success of most of these attacks mainly depends on their timeliness, as the degree of system vulnerabilities varies from time to time by changing its operating state. In this paper, we propose a new denial of service (DoS) attack strategy where the attackers leverage learning capabilities of convolutional neural networks in the encrypted domain to forecast the optimal time of launching a DoS attack. In our simulations Internet Protocol Security (IPsec) is used to secure communication channels between phasor measurement units, phasor data concentrators, and the regional/national control center. It is illustrated that, despite providing confidential communication channels by IPsec-based security gateways, an attacker still can estimate the future operating state of the power system in advance. This gives an opportunity to the attackers for initiating an effective DoS attack. The proposed method is validated by the simulation results, which show a significant increase in the success rate of DoS attacks.