A Prediction-Based Method for False Data Injection Attacks Detection in Industrial Control Systems

False data Injection attacks is an important security issue in Industrial Control Systems (ICS). Indeed, this kind of attack based on the manipulation and the transmission of corrupted sensing data, can lead to harmful consequences such as disturbing the infrastructure functioning, interrupting it or more again causing its destruction (overheating of a nuclear reactor). In this paper, we propose an unsupervised machine learning approach for false data injection attack detection. It uses a Recurrent Neural Network (RNN) for building a prediction model of expected sensing data. These latter are compared to received values and an alert security is raised if these values differ significantly.