Attack-Aware Edge MPC for Cyber-Resilient Industrial Process Control
Abstract
This paper presents an attack-aware edge model predictive control (AA-EMPC) framework. To acheive resilient process control against cyber-attacks, the proposed method augments a redundant MPC-over-edge (RMPCoE) architecture with a Secure Universal Error Function (S-UEF) and a sliding-window health index in the supervisory layer. The S-UEF generates a normalized cyber-physical risk indicator by combining local-edge state mismatch, input mismatch, output prediction error, network anomaly, and edge trust degradation. In addition, a sliding-window health index aggregates recent tracking performance, anomaly severity, and safe-operation status. These scalar indicators drive an attack-aware selection law that determines, at each sampling instant, whether the actuator should follow the local MPC, the edge MPC, or a safe command. Simulation studies on a first-order thermal system under nominal, fault, cyber-attack, and recovery cases demonstrate that AA-EMPC outperforms RMPCoE during cyber-attacks, providing a practical cyber-resilient industrial control system.