Deep Learning based Efficient Anomaly Detection for Securing Process Control Systems against Injection Attacks

Abstract

Modern Industrial Control Systems (ICS) represent a wide variety of networked infrastructure connected to physical world. Depending on the application, these control systems are termed as Process Control Systems (PCS), Supervisory Control and Data Acquisition (SCADA) systems, Distributed Control Systems (DCS) or Cyber Physical Systems (CPS). ICS are designed for reliability; but security especially against cyber threats, is also a critical need. In particular, an intruder can inject false data to disrupt the system operation. Existing approaches are not satisfactory as novel attacks against critical industries are frequently identified. Despite significant effort and progress, attacks continue to evolve. Anomaly-based detection approaches are used to detect attacks that features the injection of spurious measurement data and proven to be efficient. In this paper, we develop injection attack detection system that uses deep learning algorithms such as stacked auto encoders and deep belief networks that are tailored to identify different types of injection attacks. A model plant is used to obtain different data such as sensor and actuator measurements and specific attacks were injected into the data. The injected attacks vary in behaviour for training and testing of the proposed schema. Performance metrics to evaluate the efficiency of the proposed anomaly detection approach is presented.