Forgery Cyber-Attack Supported by LSTM Neural Network: An Experimental Case Study.

Area of Science:

• Cybersecurity
• Control Systems Engineering
• Artificial Intelligence

Background:

• Industrial control systems (ICS) are increasingly vulnerable to cyber-attacks.
• Attacks can cause significant damage to physical processes.
• Long Short-Term Memory (LSTM) networks can model complex system dynamics.

Purpose of the Study:

• To investigate the vulnerability of network industrial control systems to cyber-attacks.
• To explore the use of LSTM networks in modeling and potentially exploiting these systems.
• To propose protection methods against LSTM-based cyber-attacks.

Main Methods:

• Experimental studies conducted on a magnetic levitation process within an industrial control network.
• Development and evaluation of LSTM neural network models.
• Model training, performance evaluation, and structure selection for LSTM networks.

Main Results:

• The chosen LSTM network accurately mimicked the dynamics of the magnetic levitation process.
• Demonstrated the capability of LSTM networks to model and potentially mislead process operators.
• Identified specific cyber-attack vectors using LSTM models.

Conclusions:

• LSTM networks present a viable method for attackers to compromise ICS, even without full knowledge of the physical process.
• The study provides a foundation for developing effective defense strategies against such advanced cyber threats.
• Recommendations for protection methods against LSTM-based cyber-attacks were formulated.