Iterartive control systems design based on experimental data

通过发展前一年的(I)用于鲁棒控制的系统辨识和(Ii)用于数值优化的结果，在本年度取得了以下结果：*1.用于鲁棒控制的系统辨识*基于前一年的子空间状态空间辨识的结果，获得了系统辨识和控制系统设计的统一方法。由于鲁棒控制需要一组摄动模型，我们提出了一种识别这类模型集的方法，通过考虑附加噪声的统计特性来降低保守性。2.通过数值优化设计控制系统*我们通过参数依赖的线性矩阵不等式(LMI)开发了一种在鲁棒控制中起重要作用的u分析和综合方法来处理物理参数摄动。研究了一种直接由输入输出数据构造控制系统的方法，并提出了一种基于凸优化的学习方法。*3.对前一年制造的4输入6输出直升机系统的控制系统辨识进行了实验，将其划分为多个子系统。在进行了H*控制的基础实验后，考虑了各子系统之间的干扰，重新设计了控制器，并在圆轨迹悬停跟踪实验中取得了较好的效果。然而，由于传感器精度不高等原因，系统辨识和控制系统设计没有达到统一的结果。

The following results have been achieved in the current year by developping the results of(i)System indentification for robust control and(ii)Numerical optimization which were done in the previous year.*1.System identification for robust control*A unified approach of system identification and design of control systems has been obtained based on the results on subspace state-space identification which were done in the previous year. Since robust control needs a set of perturbed models, we proposed one method of identifying such a set which reduces the conservativeness by considering statistical properties of the added noise.*2.Design of control systems via numerical optimization*We exploit a method of mu analysis and synthesis, which play an important role in the robust control to handle physical parameter perturbation, via parameter dependent LMI(linear matrix inequality). It is expected that the proposed method will improve the performance of the control systems.A study on a method of constructing a control system directly from input-output data without the knowledge plant model are discussed, and one such learning method by convex optimization is proposed.*3.Experiment of controlling a helicopter*System identification of the 4-input 6-output helicopter system, which was manufactured in the previous year, have been done by dividing it into for sub-systems. After a basic experiment via H_* control, we re-design a controller by considering the interference among the sub-systems, and then we have obtained a good result in an experiment of havering and tacking of a circle trajectory. However, the unified result of system identification and design of control systems were not achieved because of insufficiency of the sensor precision etc.