Control-Oriented System Identification

9509539池拟议研究的目标是开发计算工具，以弥合经典系统辨识和稳健控制之间的差距，并了解系统辨识的基本限制，特别是在反馈控制的背景下。拟议的研究将使控制和识别之间的联系更加透明，并允许用户在识别方法、实验设计和模型验证方面做出智能决策。该项目将在将学习和信息论与辨识、一般的蒙特卡罗方法以及时变系统的自适应调节联系起来方面产生影响。这些主题包括模型验证的确定性方法、样本数据模型验证、模型验证的统计方法、从输入输出数据生成稳健控制相容不确定性模型的方法以及确定由时变和噪声施加的系统辨识的基本极限。***

9509539 Poolla The objective of proposed research is to develop computational tools that bridge the gap between classical system identification and robust control, and to understand the fundamental limits of system identification, particularly in the context of feedback control. The proposed research will make the connection between control and identification more transparent, and allow users to make intelligent decisions on identification methodologies, experiment designs, and model validation . The project will have implications in connecting learning and information theory to identification, in general Monte Carlo methods, and in adaptive regulation of time-varying systems. The topics include deterministic methods for model validation, sample-data model validation, statistical methods for model validation, methods for generating robust control compatible uncertainty models from input-output data, and determining fundamental limits of system identification imposed by time-variation and noise. ***