基于机器学习正则化理论的永磁同步电机转矩跟踪型MTPA控制方法  

Torque-tracking MTPA control strategy of permanent magnet synchronous motors based on machine learning regularization theory

作　　者：漆星[1] 郑常宝[1] 曹文平 张倩[1] QI Xing;ZHENG Changbao;CAO Wenping;ZHANG Qian(College of Electrical Engineering,Anhui University,Hefei 230601,China)

机构地区：[1]安徽大学电气学院,安徽合肥230601

出　　处：《电机与控制学报》2023年第11期138-148,共11页Electric Machines and Control

基　　金：国家自然科学基金(51507001)。

摘　　要：内置式永磁同步电机(IPMSM)中的最大转矩电流比控制(MTPA)是交流电机控制中的经典问题。电动汽车用IPMSM要求其控制策略不仅能够满足MTPA,还能够精确地跟踪转矩指令。为解决这一问题,引入机器学习中的正则化理论,将传统的MTPA控制问题转化成机器学习中的L1、L2正则化问题进行求解。首先将MTPA控制问题等效为机器学习中的L2正则问题,再对上述L2正则问题中的转矩约束条件进行L1正则转矩建模,从而实现对IPMSM的转矩跟踪;最后使用拉格朗日对偶方法,对正则化后的MTPA问题进行最优化求解。理论分析和试验结果表明,将IPMSM中的MTPA控制问题转化为正则化问题求解后,可以得到兼顾最大转矩电流比和高转矩跟踪精度的最优电流分配方案。所提方法结构简单、易于解释,还可以避免由于模型误差和电感饱和特性而造成的性能降低,从而融合了模型驱动法和数据驱动法的优势。Maximum torque per ampere(MTPA)in internal permanent magnet synchronous motor(IPMSM)is a classical problem in AC motor control.The control strategy of IPMSM for electric vehicle not only need to achieve the MTPA,but also need to accurately track the torque commands.In order to solve above problem,a regularization concept from machine learning theory was introduced to transform the traditional MTPA problem into L1 and L2 regularization issues.Firstly,the MTPA control problem is equivalent to the L2 regularization issue,and then the L1 regularization torque modeling was carried out for the torque tracking.Finally,the Lagrange dual method was used to optimize the above regularization-based MTPA problem.Theoretical and experimental analysis show that the proposed method can achieve an optimal current distribution scheme which both the maximum torque current ratio and the high torque tracking accuracy can be considered.Moreover,the proposed method solves the problem in a simple and analytical manner,and the solution is easy to be interpreted.Thus,it combines the advantages of model-driven and data-driven methods.

关 键 词：内置式永磁同步电机 最大转矩电流比 转矩跟踪 机器学习 正则化 拉格朗日对偶 

分 类 号：TM351[电气工程—电机]