Research On Data-based Intelligent Control Algorithm For Nonlinear Systems

The data-driven control strategy for nonlinear systems is one of hot research topics in the intelligent control fields.Considering the ubiquitous strong nonlinear links,timevarying system parameters/structure,and uncertainty of interference in the actual systems,the system models established by mathematical methods often have unmodeled dynamic,which leads to that the control methods based on system models are limited in the idea of solving the control problems of actual nonlinear systems.Therefore,it is of great theoretical and practical significance to explore intelligent control algorithm that directly utilizes the nonlinear systems input and output data.The system controller designed in this thesis is directly based on the online observable input and output data of the actual systems without depending on the systems mathematical models.The main contents of this thesis are as follows:(1)A data-driven control algorithm named Levenberg·Marquardt-PID Neural Network-Relay Feedback(LM-PIDNN-RF)algorithm is proposed for the feedback tracking control of nonlinear systems with unknown dynamics.The algorithm can achieve the feedback tracking control of nonlinear systems only by using the online control input and system output data.Firstly,relay feedback is used to stimulate the unknown controlled system to generate data.On the one hand,it solves the initialization problem of control parameters,and on the other hand,it provides a basis for the neural network identifier to approach unknown controlled system.Then,the PID neural network controller is designed based on the estimated pseudo-partial derivatives of system to achieve dynamic target tracking control of unknown nonlinear systems.The stability of the closed-loop system with the propsoed algorithm is proved by Lyapunov stability theory.In this chapter,numerical simulations and the algorithm comparisons are carried out on continuous stirring reactor,time-varying parameter nonlinear system and Hammerstein nonlinear system,the results verify the superiority and effectiveness of the proposed algorithm.The physical DC/DC Buck circuit experiments are carried out to verify the reliability of the algorithm.(2)In order to further accelerate the response speed of nonlinear systems with unknown dynamics,a dynamic autotune control algorithm(DATCA)is designed based on the LM-PIDNN-RF algorithm.By introducing a learning factor(dynamic scaling factor)adjusted exponentially,the system control input can be quickly and flexibly updated.In addition,as a data-driven control algorithm,DATCA requires one parameter set manually,which improves the applicability of the data-driven control algorithm.Besides,the complete stability proof of closed-loop nonlinear system with DATCA is proven.The large-delay system and Tianjin University’s underwater glider Petrel 200 are as examples,comparing with relay feedback PID algorithm,model-free adaptive control algorithm(MFAC),and LM-PIDNN-RF algorithm,the simulation results verify effectiveness and superiority of the DATCA.(3)Based on the idea of DATCA,PIDNN-Cohen-Coon(PIDNN-CC)algorithm is designed for solving the problems of general PIDNN control parameter initialization and closed-loop stability proof.Cohen-Coon algorithm is used to solve the general PID neural network control parameter initialization.The Lyapunov stability theory is used to prove the closed-loop stability of the general PIDNN control algorithm.The given stability sufficient conditions are consistent with the existing known conclusions.The simulation examples verify the effectiveness of the proposed algorithm.(4)Based on PIDNN-CC algorithm,a real-time intelligent control algorithm based on PID neural network-dynamic linearization identification(PIDNN-DLI)is proposed for feedback tracking control problem of unknown nonlinear systems with real-time requirements,where DLI is used to establish virtual data model equivalent to the nonlinear system to obtain the pseudo partial derivative of the system in real time.The closed loop stability of the PIDNN-DLI is proved by Lyapunov stability theory.In this chapter,the nonlinear time-varying structure system and non-minimum phase nonlinear time-varying structure system are used as examples with MFAC and LM-PIDNN-RF algorithm,numerical results verify the effectiveness and superiority of the PIDNN-DLI algorithm.PIDNN-DLI algorithm has good real-time performance via the analysis of algorithm complexity.