基于数据驱动的永磁同步电机深度神经网络控制  

Deep neural network control for PMSM based on data drive

作　　者：李耀华[1] 赵承辉 周逸凡[1] 秦玉贵 LI Yao-hua;ZHAO Cheng-hui;ZHOU Yi-fan;QIN Yu-gui(School of Automobile,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学汽车学院,陕西西安710064

出　　处：《电机与控制学报》2022年第1期115-125,共11页Electric Machines and Control

基　　金：国家自然科学基金(51207012);陕西省自然科学基金(2021JM-163)。

摘　　要：针对有限集模型预测转矩控制(MPTC)计算负担大导致实时性较差的问题,提出了基于数据驱动的永磁同步电机深度神经网络(DNN)控制方法。通过训练深度神经网络,使其学习并逼近MPTC的电压矢量选择规律,从而取代MPTC进行电压矢量的选择。通过扩充动态数据集,解决因动静态数据失衡引起的系统失控问题。通过更换训练数据集,基于数据驱动的DNN可学习并实现带非线性约束环节的MPTC。仿真验证了基于数据驱动的永磁同步电机神经网络控制的可行性,电机系统运行良好,可实现四象限运行,稳态控制效果与MPTC基本相当。Aiming at the problem of bad real-time performance due to the large calculation of the finite control set model predictive torque control(MPTC),the deep neural network(DNN)control for permanent magnet synchronous motors(PMSM)based on date drive was proposed.The data produced by the PMSM-MPTC system were used to train a deep neural network.The DNN can learn the rule to select optimal voltage vectors in MPTC and be used to control the PMSM instead to MPTC.The PMSM-DNN system may run out of control as the imbalance between the steady data and dynamic data.By extending dynamic data and balancing the data set,the problem was solved.And the DNN can learn the rule of MPTC with nonlinear constrain term only by change the training data.Simulation results testify effectiveness of the DNN for PMSM based on data drive.The PMSM under the control of the DNN works properly and operates in four quadrants.The control performances of the DNN are almost the same as the MPTC.

关 键 词：永磁同步电机 模型预测转矩控制 数据驱动 深度神经网络 转矩脉动 磁链脉动 

分 类 号：TM351[电气工程—电机]