迭代学习控制有限精度下优化设计与实现及其应用研究

项目名称： 迭代学习控制有限精度下优化设计与实现及其应用研究

项目编号： No.61473262

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 何熊熊

作者单位： 浙江工业大学

项目金额： 84万元

中文摘要： 迭代学习控制(ILC）经过近四十年的发展，已成为重要的现代控制技术之一。然而，对非重复性噪声环境下相关的ILC理论还有待完善，ILC有限精度实现的研究还尚未展开。本课题以全局优化思想为指导，研究广义迭代学习控制（GILCS）中学习滤波器最优设计理论；以系统结构理论为基础，研究迭代学习滤波器有限精度最佳实现结构；作为理论与应用研究的结合，将ILC理论及优化实现方法应用在旋转机械的主动噪声控制(ANC)，探讨新型ANC系统的设计及实现。主要内容包括：研究在（非重复性）干扰下广义迭代学习控制的跟踪性能及收敛性,提出学习滤波器优化设计准则及优化算法；分析学习滤波器实现量化误差对广义迭代学习控制性能的影响，研究基于模块简洁系统结构集合理论及优化方法，寻找学习滤波器高鲁棒性简洁实现结构；提出基于GILCS和反馈控制复合型的ANC系统的整体优化设计思想及实现方法，为未来新型ANC系统开发提供重要理论依据

中文关键词： 迭代学习控制；学习滤波器；非重复性噪声；量化误差；有限精度实现

英文摘要： As one of the important modern control techniques, iterative learning control (ILC) has been well established during the last four decades. However, it is far away being complete as the study of ILC in the presence of non- repetitive noises still needs more efforts and research in-depth, and the issues regarding the effects of finite precision implementation of ILC algorithms have been rarely investigated. The objectives of this project are i) to investigate the optimal design of learning filters in a generalized ILC scheme (GILCS); ii) based on system structure theory, to study the optimal realization of the learning filters for finite precision implementation; 3) to apply the theories developed to active noise control (ANC) systems. The specific topics include: to investigate the tractability and convergence of the GILCs in the presence of non-repetitive disturbances and hence to propose new design criteria for learning filter design and the corresponding optimization algorithms; to analyze the effects of quantization errors which occur in the learning filters on an GILCs and to study the module-based sparse system structures and optimization techniques, which are used to search for robust and sparse learning filter structures; to investigate the application in ANCs, where a novel scheme, which combines the developed GILC theory and the classical feedback control, will be derived.

英文关键词： iterative learning control;learning filter;non-repetitive noise;quantization errors;finite precision implementation