基于鲁棒优化的系统辨识算法研究  

Research on Algorithm for System Identification Based on Robust Optimization

作　　者：钱富才[1,2] 黄姣茹[1] 秦新强[3] 

机构地区：[1]西安理工大学自动化与信息工程学院,西安710048 [2]西安工业大学新型网络与检测控制国家地方联合工程实验室,西安710032 [3]西安理工大学理学院,西安710054

出　　处：《自动化学报》2014年第5期988-993,共6页Acta Automatica Sinica

基　　金：国家自然科学基金(61273127);高等学校博士学科点专项科研基金(20116118110008)资助~~

摘　　要：输入-输出数据是解决系统辨识问题的关键要素,传统的辨识理论除了假定影响输入-输出数据干扰的密度函数已知外,还要假定输入-输出数据能够精确获得,完全忽略了所用数据的质量.本文突破了传统理论的两个假设,首先用工程上易于获得的干扰的有界集合代替干扰的密度函数,并在特定数据不确定性结构下,考虑了数据质量问题,然后,以半定规划为基础,导出了鲁棒对等式,从而将系统辨识转化为对数据质量具有鲁棒性的优化问题,通过求解该优化问题,得到了一种新的鲁棒优化辨识方法,仿真结果表明了新方法的可行性和有效性.Input-output data is a key element in solving the problem of system identification. The traditional identification theory takes into account the assumptions that the density func- tion of the disturbance is known and the input-output data can be accurately obtained, while completely ignores the quality of the data used. In the paper, to overcom the limitation of the two assumptions, a bounded set is firstly taken which can be ob- tained easily in engineering as an alternative to the density func- tion. Subsequently, with the specific uncertain data structure and considering the effect of the data quality, robust counterpart is derived by the semi-definite programming theory. And, the system identification problem is converted to an optimization problem which is robust to the uncertain data. By solving the optimization problem, a new identification algorithm based on robust optimization is proposed. Simulation results show the feasibility and effectiveness.

关 键 词：系统辨识 不确定性 鲁棒优化 半定规划 

分 类 号：N945.14[自然科学总论—系统科学] TP13[自动化与计算机技术—控制理论与控制工程]