Machine Learning-Based Cyber-attack Detection and Resilient Operation via Economic Model Predictive Control for Nonlinear Processes

Abstract

This work proposes resilient operation strategies for nonlinear processes that are vulnerable to targeted cyber-attacks, as well as detection and handling of standard types of cyber-attacks. Working with a general class of nonlinear systems, a modified Lyapunov-based Economic Model Predictive Controller (LEMPC) using combined closed-loop and open-loop control action implementation schemes is proposed to optimize economic benefits in a time-varying manner while maintaining closed-loop process stability. Although sensor measurements may be vulnerable to cyber-attacks, the proposed controller design and operation strategy ensure that the process will maintain stability and stay resilient against particular types of destabilizing cyber-attacks. Data-based cyber-attack detectors are developed using sensor data via machine-learning methods, and these detectors are periodically activated and applied online in the context of process operation. Using a continuously stirred tank reactor example, simulation results demonstrate the effectiveness of the resilient control and detection strategy in maintaining stable and economically optimal operation in the presence of cyber-attacks.