Iterative Learning Control And Optimization Under Active-type Incomplete Data

To achieve perfect control performance,the existing results of iterative learning control(ILC)mostly depend on the acquisition and utilization of complete system information and operation data.Provided that the tracking precision and control performance is acceptable,it is important to reduce the acquisition and computation of data due to several considerations.Specifically,the reduction can lower the hardware and software cost,improve the system efficiency,and enhance the system robustness largely.Therefore,it is of great theory significance and application value to study how to design data-driven ILC to achieve high quality tracking performance under incomplete data environments.This thesis considers the robustness of ILC and the corresponding control system design based on active-type incomplete data.Here,the active-type incomplete data means the incomplete data and information caused by man-made reduction of data quantity and quality on the premise that the given objective is achieved,such as sampling,quantization and point-to-point(P2P)control.The main results are as follows:1.Chapter 2 studies the sampled-data ILC(SDILC)problem.We first present an evaluation of the upper bound of the interval tracking error and then propose an varying-sampling technique to enhance the interval tracking performance.The upper bound estimation of the interval tracking error shows that the higher the sampling rate or the smaller the sample interval,the smaller the upper bound of interval tracking error,and thus the better the tracking performance.The proposed varying-sampling strategy includes the following steps.For a continuous system and arbitrary initial sampling frequency,first establish a sampled-data ILC ensuring that the at-sample tracking error converges into a desired range.Then check the interval behavior to determine which intervals are not acceptable in the sense that the interval tracking error exceeds the desired bound.After that,increase the sampling frequency for the unacceptable intervals.By repeating the above steps,the tracking error at the sampling instants and intervals can both asymptotically converge into the desired bound.2.Chapter 3 studies the quantized ILC for deterministic systems and the influence of logarithmic quantization on ILC design and analysis.An ILC update framework based on quantized error information is also proposed.In the proposed framework,the tracking error rather than the output is quantized and utilized for updating the system input signal.The tracking error is proved asymptotically convergent to zero.3.Chapter 4 considers the quantized ILC problem for stochastic systems.The ILC framework in Chapter 3 is applied with an introduction of a decreasing learning gain sequence along the iteration axis.The decreasing sequence is used to eliminate the effect of stochastic noises and quantization errors and ensure a stable convergence of the input sequence.It is proved that the input sequence converges to the optimal input with probability 1.4.Chapter 5 studies the point-to-point tracking problem for stochastic systems and proposes an ILC update law based on stochastic approximation.In this chapter,we first define a matrix to connect the required tracking positions and the original tracking trajectory.Discussions on the design of the learning gain matrix are also presented.A rigorous analysis on the convergence and asymptotically statistical property of the proposed ILC algorithm is given.