【目的】针对同步磁阻电机(SynRM)驱动系统运行过程中磁饱和效应引起的d-q轴电感非线性畸变问题,提出一种基于鲁棒递推最小二乘法(RRLS)的在线电感参数辨识策略。【方法】首先,计算预测电压差构建历史预测残差序列,在电机运行过程中滚动优化,有效降低偶然数据导致的稳态估计误差。其次,使用预测标准差作为鲁棒尺度来构建鲁棒损失函数,增强算法抗负载扰动能力,且未显著增大计算量。然后,利用近似平衡条件结合带可变遗忘因子的自适应机制递推,多次迭代得到准确的参数估计值。最后,在Matlab/Simulink中搭建SynRM控制及参数辨识系统,并在不同运行条件下,将RRLS算法与传统变遗忘因子递推最小二乘法(VFFRLS)进行对比分析。【结果】仿真结果表明,在空载与负载扰动条件下,所提RRLS算法具有更低的辨识误差,d轴电感稳态误差小于0.5%,q轴电感稳态误差小于4%。在动态过程中,所提RRLS算法将d轴超调量由VFFRLS算法的25 mH降至12 mH,将q轴超调量由VFFRLS算法的33 mH降低至13 mH。【结论】与传统VFFRLS算法相比,本文所提RRLS算法的电机参数在线辨识方法可实现稳态高辨识精度,降低动态过程中超调量,且在负载扰动下在线辨识结果良好,系统鲁棒性高。

[Objective] Aiming at the nonlinear distortion of d-q axis inductance caused by magnetic saturation effects in the operation of synchronous reluctance motor (SynRM) drive systems, this paper proposes an online inductance parameter identification strategy based on robust recursive least square (RRLS). [Methods] Firstly, the predicted voltage difference was calculated to construct a historical prediction residual sequence, and rolling optimization was performed during motor operation to effectively reduce steady-state estimation errors caused by random data. Secondly, the predicted standard deviation was used as a robust scale to construct a robust loss function, which enhanced the algorithm’s ability to resist load disturbances without significantly increasing the computational burden. Then, an approximate equilibrium condition was combined with an adaptive mechanism with a variable forgetting factor for recursive estimation, and accurate parameter values were obtained through multiple iterations. Finally, a SynRM control and parameter identification system was built in Matlab/Simulink, and the RRLS algorithm was compared with the traditional variable forgetting factor recursive least square (VFFRLS) under different operating conditions. [Results] The simulation results showed that under no-load and load disturbance conditions, the proposed RRLS algorithm had lower identification errors. The steady-state error of the d-axis inductance was less than 0.5%, and the steady-state error of the q-axis inductance was less than 4%. During the dynamic process, the d-axis overshoot was reduced from 25 mH by the VFFRLS algorithm to 12 mH by the proposed RRLS algorithm, and the q-axis overshoot was reduced from 33 mH to 13 mH. [Conclusion] Compared with the traditional VFFRLS algorithm, the RRLS algorithm proposed in this paper achieves high steady-state identification accuracy, reduces overshoot during dynamic processes, and demonstrates excellent online identification performance under load disturbances, with high system robustness.

中国创新挑战赛(宁波)重大专项(2024T004)；武汉市科技局重点研发项目(2024060702030146)