Stealthy cyberattacks on control systems using an adaptive soft-constrained optimization method

This paper presents a novel approach for designing stealthy cyberattacks on automated control systems of critical infrastructure. The core idea lies in employing an adaptive soft-constrained optimization method, which simultaneously maximizes the impact functional of the attacker while keeping the attacked trajectory within the invisibility range of a standard fault detection mechanism. The proposed approach is based on a variational problem formulation, the construction of adjoint equations, and a gradient-based procedure with dynamic penalty parameter updates. Numerical simulation is conducted on a second-order test dynamic system. The results demonstrate the algorithm's effectiveness and convergence, as well as the feasibility of generating a controlled attack that successfully bypasses WLS-based detection methods. The method can be used to test the resilience of industrial systems to cyber threats through security scenario modeling.