Research On Neural Network Embedded Learning Control

With the development of control theory,the problem of high control performance of nonlinear systems has attracted widespread attention.In recent years,the large number of applications of intelligent robots such as aerial work robots and upper limb rehabilitation robots have led to higher requirements for the control performance of nonlinear systems,but the control objectives of high precision,good stability and fast response speed can no longer be achieved using classical control theory.Based on the Lyapunov system algorithm,a neural network controller is introduced to establish a loss function targeting the tracking error,thus further reducing the tracking error and improving the control performance on top of the tracking error of the original controller with great practical significance.In this thesis,learning control methods for nonlinear systems and their applications are investigated for nonlinear systems containing unknown nonlinear terms as well as perturbations,The main work and innovation of this paper are as follows:(1)For a class of nonlinear systems containing unknown parameters and external disturbances,the mathematical model of this system is first established,and the control scheme combining neural network and sliding mode control techniques is used for this system to achieve improved control performance,which is verified by simulation of the first-order-second-order as well as motor system.The simulation results show that the effectiveness of the control algorithm is verified.It is also compared with the current advanced control algorithms to further verify the advancedness of the control algorithm.(2)A neural network embedding-based control scheme for the flight platform is proposed for the high control performance problem of the fully driven air operation system flight platform.First,for the flight platform attitude system,the backstepping controller is used as the base controller,and the hybrid basis function and improved disturbance observer are used to compensate for the model uncertainty and disturbance problems,respectively.Secondly,for the platform position system,the linear active disturbance rejection controller is used as the base controller,and the linear expansion state observer technique is used to estimate the "total disturbance" for the model uncertainty and disturbance problem of the position system.At the same time,a neural network controller with tracking error as the objective function is embedded on top of the basic controller to improve the control performance of the basic controller.Finally,virtual experiments are conducted in Coppelia Sim software and advanced control algorithms are compared to verify the effectiveness and advancement of the learning controller.The proposed neural network learning control method can ensure that the designed control scheme has smaller tracking error than that before adding the neural network controller when the system model is not fully known and there are external disturbances.The stability analysis shows that the proposed control strategy can guarantee the stability of the closed-loop system,and the effectiveness of the proposed control method is verified by simulation examples.