电机动态参数的鲁棒最小二乘辨识方法研究

Robust Least Squares Estimation Method for Motor Dynamic Parameter Identification

最小二乘常用于辨识电机参数的动态变化,对于电机的精确控制至关重要。工况条件下的电流、电压采样都包含大量噪声,电机模型中的求导操作进一步放大了噪声,导致最小二乘的信息矩阵奇异,辨识结果出现很大偏差。为提高辨识方法的噪声鲁棒性,提出了两种思路。一是用三次样条拟合采样信号,然后求解其一阶和二阶导数,可防止高频噪声在求导过程中被继续放大。二是采用岭回归估计取代最小二乘估计,避免矩阵求逆出现病态结果,消除信息矩阵奇异性的影响。基于半实物仿真的实测数据表明,两种改进方法均可获得更加合理可靠的辨识结果。

Least square is prevalent in motor parameter identification,which is crucial to motor＇s accurate control.The electrical current and volt samples in working motors are contaminated by noise,and the contamination is amplified by differential operations during motor modeling.The noise contamination leads to a singular information matrix and unsteady estimations.Two methods are presented in this paper to improve the noise robustness of motor parameter identification.The cube spline is used to fit sample signal,improve the smoothness of differential operations and prevent the noise contamination from being amplified.Ridge regression is used to replace the common least square method,avoid the ill-posed inverse matrix and remove the negative influence of matrix singularity.The half-physical simulation proves that the identification results with the above new methods are effective.