带有色观测噪声的系统状态参数鲁棒估计  

Robust State Estimation for Systems with Colored Measurement Noise

作　　者：柳毅[1,2] 赵振宇[1,2] 丁全心[1,2] 李建勋[3] 

机构地区：[1]中航工业洛阳电光设备研究所,洛阳471009 [2]光电控制技术重点实验室,洛阳471009 [3]上海交通大学,上海200240

出　　处：《系统仿真学报》2013年第2期293-295,300,共4页Journal of System Simulation

基　　金：重点实验室基金(9140C4602041001;9140C460104110C4602);专用技术研究项目(402040103)

摘　　要：针对非平稳观测系统的噪声,考虑了白噪、色噪声和尖点噪声同时存在的情况,通过变形转化,将系统噪声表示成白噪的形式,首次给出了统一的观测方程。转化后的观测方程在色噪声ARMA辨识基础上,可以直接用于传统的卡尔曼滤波算法,避免了扩维滤波。于是状态参数的稳健估计归结为ARMA参数的辨识问题。通过鲁棒支持向量回归机辨识ARMA参数,根据噪声特性选择合适的损失函数自由参数,从而实现了ARMA参数的鲁棒辨识。In practical application, the observation systems was influenced by white noise, colored noise and outliers. With transformation, a uniform measurement equation was derived by denoting the colored noise with white noise. Kalman filtering model for colored noise was proposed in order to avoid complicated computation and expansion of the dimension of the filter. The parameters of ARMA model were estimated by robust support vector regression （SVR）, hence, the estimation error of ARMA model that resulted from noise and outliers was greatly decreased. Based on Kalman filtering model for colored noise and robust ARMA identification, robust state estimation for systems with unstable and colored measurement noise could be handled.

关 键 词：色噪声 鲁棒支持向量机 ARMA模型 鲁棒辨识 

分 类 号：TN911.7[电子电信—通信与信息系统]