Iterative Learning Control Methods For Two Types Of Non-repetitive Systems

Iterative Learning Control(ILC)is an emerging research direction in the field of modern automatic control,which effectively improves the transient performance in the time domain by utilizing the repeatability of the systems when the same control task is carried out.In practical engineering,however the strict assumption on repeatability always fails to be satisfied.Recently,study on non-repetitiveness by relaxing the strict assumption becomes a research focus in the field of ILC,and it is also considered as an important scientific problem of ILC during its progress of filling the gap between theory and application.In this thesis,two iteration-varying problems are discussed for two types of non-repetitive discrete-time systems,respectively,and ILC methods are proposed correspondingly.In particular,a novel designing methodology of integrating classification algorithm with iterative learning controller is first proposed.The main works and contributions are summarized as follows.1.For a type of linear discrete-time systems,the issue of iteration-varying system parameter which varies among different categories at different iterations is first studied.By introducing this issue,the research content on non-repetitiveness is enriched.On the one hand,the allowable variation range of the system parameter in the iteration domain is extended to a great degree when this issue is considered.On the other hand,the category characteristics of the system parameter caused by its variation in the iterative domain is fully considered from a practical point of view.2.A novel ILC algorithm based on classifier is proposed for the linear discrete-time systems with non-repetitive parameter which varies among different categories.With the help of the trained classifier,the class of the non-repetitive parameter is labeled according to real-time input and output data.ILC controller is updated using the historical data in the corresponding category.The bounded convergence of the proposed algorithm without classification errors is proved through theoretical analysis.Compared with the traditional ILC algorithm,the proposed one has obvious advantages.3.For the real-time requirement and data storage problem when iterative learning control is integrated with classifier,the class rectifying mechanism and data saving mechanism are studied,respectively.On the one hand,two kinds of class rectifying mechanisms based on voting method are proposed to correct the real-time system labels obtained by the classifier.On the other hand,the idea on data saving in category for ILC is put forward.New real-time and batch-wise data saving methods in category is developed,which save historical data in different memories according to the label of system parameter.The characteristics of proposed class rectifying mechanisms and data saving mechanisms are analyzed,and the effectiveness when they are applied to ILC integrated with classifier is verified by a plenty of simulation results.4.For the macroscopic freeway traffic system,which is a typical distributed parameter system,a ramp metering strategy based on feedback-aided ILC method is designed,in the presence of iteration-varying boundary conditions.To deal with the non-repetitiveness caused by different downstream traffic conditions in the section closed to the boundary,the real-time feedback module utilizes the density tracking error in boundary section.Meanwhile,the repetitive characteristics of traffic flow is caught and learned by ILC module.By rigorous analysis,the bounded tracking performance of the density error is proved when the proposed ILC algorithms is applied to dealing with non-repetitive boundary conditions.The effectiveness of the proposed feedback-aided ILC schemes for distributed parameter systems with non-repetitive boundary conditions are verified by simulation.