Intelligent Model‐Free Data‐Driven Robust Control Based on Fractional Reinforcement Learning

ABSTRACT Precise modeling of real complex nonlinear systems is fundamental to control design, but reaching it in practical engineering and industrial applications is often challenging. Data‐driven approaches, by effectively capturing complex nonlinear behaviors, can enhance the performance of nonlinear robust controllers. In this paper, a new intelligent discrete fractional model‐free backstepping sliding mode controller is proposed to enhance the performance and stability of practical engineering systems by utilizing a reinforcement learning (RL)‐based disturbance estimator. The proposed controller leverages the advantages of Caputo fractional‐order discrete sliding mode control, offering higher adaptability in handling uncertainties and external disturbances. Additionally, a Caputo fractional‐order RL‐based disturbance estimator is designed to estimate and compensate for the existing disturbances, without requiring an accurate model of the system, which makes it more applicable for various real complex systems. All controller coefficients in this study have been optimally selected using the Ant Colony Optimization algorithm. Its performance is evaluated through simulations, comparing it to the conventional discrete Backstepping Sliding Mode Controller (DBSMC). The proposed intelligent model‐free controller, due to its data‐driven and model‐free structure, has the potential to be applied to various practical engineering systems. Its effectiveness is demonstrated through a wind turbine case study under different operating scenarios. In this study, the wind turbine system under three scenarios: normal conditions, parametric uncertainty, and external disturbances are examined. The results demonstrate that the proposed method outperforms the DBSMC, with improvements of 69.25% in normal conditions, 65.28% under parametric uncertainty, and 76.23% under external disturbances. These findings highlight the superior performance and robustness of the proposed approach in various operating conditions.