Research For A Class Of Nonlinear Systems Iterative Learning Control Methods

Iterative learning control is a branch of the intelligent control, and it has the "learning"capability. By learning the previous control experiences, in the control process it can correctthe control input signals in the next time, and Iterative learning control is based on the controlerrors, rather than mathematical model of the controlled object. Therefore it is suitable for thelinear time-invariant systems and the complex nonlinear time-varying systems. The morecomplex and uncertain the model is, the better iterative learning control will be. But thetraditional iterative learning control always uses the fixed parameters learning law, and thecontrol convergence speed is limited, inflexible, and dynamic process exists oscillation. Thisarticle will make the research on improving a class of nonlinear repetitive motion systemsiterative learning control convergence speed and the dynamic process of the oscillationphenomenon, and further for researching on how to improve iterative learning controlmethods to control the initial state errors and input delay.Firstly, this paper studies a class of nonlinear repetitive system iterative learning controlmethod, to analyze learning gain parameters affect on the control process. Then an improvedmethod is raised which uses fuzzy controller design learning gain. According to the size of thelearning error fuzzy controller adjust the learning gain. But in the design portion of the fuzzycontroller, fuzzy rule editing have a relationship with controlled object state variabledimension, too many variables can cause an explosion rules, the controller implementation isdifficult, therefore, combined with feedback linearization algorithms and information fusiontechnology, the dimension of the fuzzy controller input variables reduces, and a fuzzy iterativelearning controller design achieve. Iterative learning control compared to conventionalcontrollers multivariable problem solving, learning speed significantly improved, greatlyimproved control stability.Second, as a lot of interference for the actual controlled object, it makes the systemcontrol features decline. Such as measuring and execution equipments errors will cause theinitial state shift, and controller calculates speed and data transmission delay will cause thesystem input delay and so on. For the control errors of the object’s initial state, this researchpropose to add a control unit for study on object’s initial state, through continuous "learning"initial state error, so that the system output values and the initial value converges to thedesired value. Since the system output and the initial value learning is relatively independent,iterative learning controller convergence speed is not affected. Proven, fuzzy iterative learning controller with initial state learning can make the system fast convergence in expectations,and the dynamic characteristic curve smoothing in both cases of systems fixed and arbitraryinitial state error. Further, for the controlled object’s input delay problem, this paperintroduces an advanced ahead given algorithm to improve the controller. The learning errorcaused by input delay is given early to the iterative learning controller to eliminate controlerrors as much as possible. Operator theory proved convergence of the algorithm, in the caseof expectation initial value, the controller can make the system fast convergence.On this basis,this paper propose an error ahead fuzzy iterative learning control method which has the initialstate learning ability, it has greatly improved iterative learning controller performance of thesystems which have the initial state error and input delay. By two degrees of freedommanipulator trajectory tracking control simulation, fully demonstrate the proposed controlmethods superiority. Because in this study the control objects are closer to the actualproduction system, so the iterative learning control solutions have great significance forindustrial applications in a class of nonlinear repetitive system control problems.