Cyber-attack detection method based on RNN

Abstract

Current and forthcoming market requirements bring huge challenges to today manufacturing. Answer to the changing demands and high product variety is found in the integration of the Internet of Things (IoT) and Cyber-Physical Systems (CPS) into industrial plants. CPS as smart devices capable of data processing and information exchange enable fast adaptation of manufacturing resources to production of diversified products. Nevertheless, fully implemented internet communication at factory shop floor opens up a whole new area for potential cyber-attacks. The consequences of attacks can have a negative influence on the system or even endanger human lives. Therefore, defence techniques must be developed to ensure a high level of protection. Early detection of cyber-attacks is crucial to minimize or completely avoid the negative effects of the attack and keep the system safe and reliable. In this work, we propose an attack detection method based on deep learning approach. We explore the application of several deep learning architectures based on Simple Recurrent Neural Networks (Simple RNN) and Long Short-Term Memory (LSTM) based RNN for generation of the detection mechanisms tailored to the concrete process. Our method was experimentally verified using real world data and it proved to be effective, as it detected all considered attacks without false positives.