Communication pattern anomaly detection in process control systems

Abstract

Digital control systems are increasingly being deployed in critical infrastructure such as electric power generation and distribution. To protect these process control systems, we present a learning-based approach for detecting anomalous network traffic patterns. These anomalous patterns may correspond to attack activities such as malware propagation or denial of service. Misuse detection, the mainstream intrusion detection approach used today, typically uses attack signatures to detect known, specific attacks, but may not be effective against new or variations of known attacks. Our approach, which does not rely on attack-specific knowledge, may provide a complementary detection capability for protecting digital control systems.