AI-Enabled Dual-Rate Predictive Control With ELM-Based Learning and Kalman State Estimation

Despite extensive advances in nonlinear model predictive control and learning-based system identification, robust control of nonlinear systems under asynchronous sensing, model uncertainty, and measurement noise remains fundamentally challenging. In particular, conventional MPC frameworks rely on accurate affine or locally linearized models and single-rate measurements, which limits their applicability to practical systems exhibiting strong nonlinearities, dual-rate sensing, and time-varying dynamics. This limitation prevents many real-world cyber–physical systems from fully benefiting from MPC’s robustness and constraint-handling capabilities. This work addresses this long-standing challenge by proposing a unified dual-rate adaptive control framework that integrates online learning, state estimation, and constrained predictive control for nonlinear systems. Unlike existing learning-based MPC approaches that require offline training, fixed model structures, or heuristic gain tuning, the proposed method provides adaptive modeling and control entirely online without prior system knowledge. This work introduces three main contributions: (1) a dual-rate unscented Kalman filter capable of reliable state estimation under asynchronous and sparse measurements, (2) an online extreme learning machine–based NARX modeling framework embedded directly within the MPC prediction model, and (3) a Lyapunov-based adaptive control architecture ensuring practical closed-loop stability and robustness. The learned model is continuously updated in real time and used within a constrained nonlinear MPC formulation with adaptive steady-state target computation, enabling accurate trajectory tracking under parametric uncertainties, external disturbances, and measurement noise. Simulation results under multiple challenging scenarios demonstrate that the proposed architecture achieves superior tracking performance and robustness compared to conventional PID schemes, highlighting its effectiveness for high-precision digital control of nonlinear systems.