基于支持向量机的开关磁阻电机转子位置在线建模

Online modeling based on support vector machine for rotor position signal of switched reluctance motors

针对开关磁阻电机离线转子位置估计器在实际工程应用中可能存在预测精确度变差的问题,提出了一种基于近似支持向量回归机的在线转子位置预测模型。根据支持向量回归机在线建模的原理,充分考虑电机的实际运行状况,结合在线预测模型对收敛速度与预测精确度的要求,提出了基于灰色关联的加权分类近似支持向量回归机模型。该模型由于只需求解线性方程组,具有计算简单快速的优点。基于灰色关联度的在线动态训练样本集的选取由于考虑了样本的时间与空间特性,使得模型具有良好的泛化性。实验结果表明:只需较小的训练样本集,便可实现电机转子位置的准确估计。

For off-line rotor position estimation accuracy of switched reluctance motor （ SRM） in practical application might become poor, online rotor position prediction model based on proximal support vector regression （ PSVR） is proposed. According to the online modeling principle of SVR, fully considering the motor actual operation condition, combined with the convergence speed and prediction accuracy require-ments of online prediction model, weighted proximal SVR machine via classification （ WCPSVR） based on grey correlation is presented. This model only needs to solve linear equations, with the advantages of simple and fast calculation. Online dynamic set of training samples based on grey correlation degree was selected taking into account the samples’ characteristics of time and space, so it made the model has a good generalization. Experimental results show that this on-lining prediction model realizes accurate pre-diction of rotor position using a smaller set of training samples.