Research On Data Driven Control Of Nonlinear Systems With Unknown Parameters

Data-driven control gets rid of the dependence on the mathematical model,and only uses the input and output data of the system operation process to learn the system model,states or control information.In this thesis,combined with optimal control the-ory,an intelligent control algorithm that meets the system performance index is given.The main contents are as follows:1.Aiming at the optimal control problem of multi-input multi-output linear discrete-time system,a two-degree-of-freedom controller is designed using virtual reference feedback tuning method.Firstly,based on the open-loop data,the data-driven optimal control problem is proposed,and then the performance index of the two-degree-of-freedom controller is established,and the optimal controller is designed by minimizing the performance index through the virtual reference feedback tuning method.2.For nonlinear systems with partially unknown information and input saturation,an H_?state feedback controller is designed by using policy iteration algorithm.First-ly,the input with saturation constraints is processed by using the quasi-norm,then the policy iteration is used to solve the Hamilton-Jacobi-Isaac equation.Finally the algo-rithm is implemented by constructing an actor-critic-disturbance neural network,and the optimal controller is obtained.3.Aiming at the suboptimal control problem of locally unknown nonlinear sys-tems with unmatched disturbances,an integral sliding mode control and policy itera-tion method are used to design controller.Firstly,the nonlinear disturbance observer is used to estimate the unknown unmatched disturbances,then the integrated sliding mode dynamic surface is designed based on the disturbance estimation to suppress the disturbance.Finally the policy iteration is used to design the suboptimal control for the above equivalent sliding mode dynamic system to meet the desired performance index.4.For nonlinear systems with external disturbances,an adaptive tracking con-troller is designed.Firstly,the hysteresis quantizer and sector constraint theory are used to decompose the actuator backlash into a feasible control strategy form.Second-ly,combined the backstepping method and policy iteration,the controller is designed.Finally,the optimal controller is solved by using the actor-critic neural network.5.An optimal control algorithm based on reinforcement learning is proposed for nonlinear partial differential equation systems which mathematical models cannot be established.Firstly,the empirical eigenfunctions of the partial differential equation sys-tem are calculated by the Karhunen-Lo`eve decomposition algorithm.Secondly,the system is converted to a higher-order ordinary differential equation system by using these empirical eigenfunctions functions,and then through the singular value perturba-tion theory,higher-order ordinary differential equation system is reduced.Finally,the reinforcement learning is used to solve the optimal controller.