Real-Time Stochastic Reference Tracking Nonlinear Predictive Functional Control: A Case Study on Wind Turbines

In recent years, the development of Nonlinear Model Predictive Control (NMPC) methods has shifted focus from set-point stabilization to tracking reference trajectories. However, uncertainties in reference signals and the need for dynamic forecasting have underscored the demand for robust NMPC frameworks. To simplify analysis and ease implementation, these approaches typically assume constant forecasting of the stochastic reference, which often leads to performance degradation. In this paper, we move beyond this simplifying assumption and propose a novel nonlinear Predictive Functional Control (PFC) framework for tracking stochastic references, leveraging fundamental PFC concepts to significantly reduce the computational burden. Feasibility and convergence to a robust invariant set are guaranteed under a cost function constraint, and the implementation is carried out using a parallel Newton method. The proposed controller is designed and evaluated for a wind turbine, with stochastic wind speed as the reference. Results show superior performance over existing methods, especially in reducing computation time and improving tracking robustness. Note to Practitioners—This paper proposes a NMPC method for tracking stochastic references, applicable to industrial systems with uncertain inputs, such as wind turbines. Many NMPC methods assume that the stochastic reference remains constant, which compromises control accuracy. This research addresses this limitation by proposing an advanced nonlinear version of PFC, offering an approach that enhances tracking accuracy and reduces computational time. PFC is utilized in various industries due to its simplicity and efficiency. The proposed method, by incorporating dynamic reference forecasting and parallel Newton iterations, enables real-time implementation. Experiments on a wind turbine demonstrate that this approach increases aerodynamic power extraction and reduces mechanical stresses. Furthermore, it can be implemented with minimal modifications in other industrial systems requiring fast and accurate predictive control.