Iterative Learning Identification And Control With Applications In High-speed Train Operation Control System

In the dissertation,new iterative learning control(ILC)approaches are s-tudied for repetitive operating nonlinear systems.For affine nonlinear systems,new adap-tive ILC methods are designed by considering both state and input constraints,random initial errors,and random external disturbances all together.And for non-affine nonlinear systems,a novel data-driven predictive iterative learning control(DDPILC)method is proposed.Moreover,for repetitive operating nonlinear high-speed train systems,a novel data-driven ILC based dynamic modeling and norm optimal ILC is proposed,aiming at addressing the problem of difficult modeling,guaranteeing safe operation and high con-trol performance.Main works and contributions in the dissertation are summarized as follows.1.Adaptive iterative learning control is studied for a class of affine nonlinear sys-tems with both state and input constraints,random initial errors,and random external disturbances.For parametric uncertainties,adaptive learning law in the iteration axis is designed.For non-parametric uncertainties,radial basis function neural network(RBFN-N)is utilized to approximate the uncertainties.The weights of the neural network are updated adaptively along the iteration axis,and the approximation error is also compen-sated by estimating its upper bound along the iteration axis.Composite energy function(CEF)with a new barrier Lyapunov function,together with a projection mechanism,is used to guarantee the convergence of tracking errors while satisfy the state and input constraints automatically.2.For a class of repeatable and unknown non-affine nonlinear systems,a novel data-driven predictive iterative learning control(DDPILC)with and without considering input and output constraints are proposed.By virtue of a constructive dynamic lineariza-tion technique(DLT),the proposed control methods depend only on the measured in-put/output data without using any model information of the controlled plant.Therefore,it is a kind of data-driven control methods.Rigorous theoretical analysis shows that with random initial operation conditions,the proposed unconstrained DDPILC scheme guar-antees monotonic and pointwise convergence property while the constrained DDPILC scheme guarantees the asymptotic and pointwise convergence.3.The accurate dynamic model of the nonlinear high-speed train operating system is hard to establish since the precise calibration of time-varying or even fast time-varying model parameters is not an easy task,although the dynamic model structure can be ob-tained using the first principle knowledge.To change the traditional "modeling before control" design step,the proposed approach applies data-driven PID-type ILC at first,and at the same time,a novel iterative learning identification algorithm is proposed to identify the time-varying parameters simultaneously.The designed identification algorithm can guarantee that by choosing appropriate gain matrix,a predetermined identification preci-sion requirement can be achieved as quickly as possible.When the model identification precision achieves the predetermined identification precision requirement,based on this well identified nonlinear train model,a norm optimal ILC with consideration of security,punctuality and travelling comfort will be designed and will substitute for the original applied PID-type ILC,to improve the control performance and ensure safe operation of the high-speed train.In addition,The established system model not only can be used to achieve high control performance,but also can realize precise forecasting,monitoring,diagnosis and other functions.Simulation and experimental studies further verify the va-lidity of the proposed approach.The conclusion shows good application perspective for the proposed approach into the filed of automatic train control.