Design And Analysis On Active Disturbance Rejection Control For Nonlinear Uncertain Systems

Disturbances and uncertainties widely exist in control systems and greatly influence stability and control performance,so effectively handling the disturbance and uncertainty has become a core issue in system control.As a general and novel controller,the active disturbance rejection control(ADRC)provides an effective solution to disturbances and uncertainties.However,its control capability and processing effect still need further exploration.Therefore,this thesis conducts in-depth studies on related problems of ADRC for nonlinear uncertain systems.Firstly,a generalized reduced-order extended state observer(ESO)with broad applicability is proposed for the fundamental design problem.Then,considering the influence of uncertain control gain on ADRC performance in nonlinear systems,the importance of control gain to ADRC design and stability is revealed.So to online separately identify the uncertain control gain and disturbance,a control scheme is proposed based on the estimation of ESO.In addition,the ability of ADRC to cope with actuator saturation nonlinearity is studied,and an online parameter tuning method is developed for ADRC-based autonomous driving combined with reinforcement learning.The specific content is as follows:(1)For nonlinear uncertain systems with multi-linear outputs,a generalized reduced-order ESO(RESO)is proposed by making full use of the measurable information of the system,together with rigorous analysis of system stability.This method can reduce the design complexity and phase delay while including the existing RESO design form.The results show that the proposed method can be applied to systems with multi-state output,high-order state output,and linear combination state output and can accurately estimate unknown states.(2)For nonlinear systems with uncertain control gain,the stability of ADRC subject to disturbances and uncertainties is studied.According to the ADRC-based tracking error and estimation error,the sufficient condition for the uncertain control gain to ensure stability is analyzed,and the uniform and ultimate boundedness is obtained.Then,the global asymptotic stability and the convergence rate of the ADRC system are analyzed under a class of disturbance.The results show that the uncertain control gain plays a vital role in system stability,and both external disturbances and system parameters determine the feasible region of control parameters.(3)The online estimation of disturbance and uncertain control gain is studied considering the importance of uncertain control gain to the ADRC-based system stability.Based on the ADRC framework,applying the estimation of ESO for the total disturbance,the uncertain control gain and disturbance are online estimated respectively combined with the sign projected gradient flow and are used for feedback design.Then,the exponential and ultimate stability for the proposed method is presented.Both simulation and experiment show the effectiveness of the proposed method.(4)For the uncertain system with actuator saturation nonlinearity,the ability of ADRC on disturbance and uncertainty rejection is analyzed.Deriving the tracking error and estimation error and applying the convex hull to deal with the actuator saturation,the ellipsoid invariant set in the domain of attraction is obtained for the ADRC-based system and the convex hull of the ellipsoid is still an enlarged invariant set.Both simulations and experiments show that ADRC can cope with the actuator saturation nonlinearity with good control performance and strong robustness.(5)Aiming at the longitudinal autonomous driving control,an intelligent ADRC controller is proposed based on reinforcement learning.In this scheme,the ADRC is applied to online estimate and compensate for the total disturbance.Reinforcement learning is used to online tune the control parameters according to the external environment.And the mapping relationship between the control parameters and the errors is established for the maximal reward.The results show that the intelligent controller has strong adaptability and good robustness.