Input-To-State Stabilizing Extreme Learning Machine-Based Model Predictive Control Under Denial-Of-Service Attack

We present a novel model predictive control (MPC) strategy integrating extreme learning machine (ELM) modeling to protect discrete-time systems against denial-of-service (DoS) attacks. Our approach employs ELM to construct data-driven prediction models, significantly reducing modeling costs compared to traditional methods. To ensure closed-loop stability under DoS attacks, we develop a specialized terminal positive invariant set that handles the linearization remainder terms from ELM-represented nonlinear systems. We establish theoretical guarantees of input-to-state stability (ISS) for the closed-loop system under bounded prediction errors. Simulations on a continuous stirred tank reactor (CSTR) and implementation on a three-wheeled omnidirectional robot platform demonstrate the effectiveness of our approach while maintaining stability under sustained DoS attacks.