Title:

Temperature prediction model for permanent magnet synchronous motors based on deep learning

作者：

岑 岗, 蔡永平, 岑跃峰

岑 岗,蔡永平,岑跃峰

Author(s):

CEN Gang, CAI Yongping, CEN Yuefeng

CEN Gang, CAI Yongping, CEN Yuefeng

单位:

(浙江科技学院 信息与电子工程学院,杭州 310023)

Unit:

(School of Information and Electronic Engineering, Zhejiang University of Science and Technology, Hangzhou 310023, Zhejiang, China)

关键词:

永磁同步电机; 温度预测; 嵌套式门控循环单元; 卷积神经网络

永磁同步电机; 温度预测; 嵌套式门控循环单元; 卷积神经网络

Keywords:

permanent magnet synchronous motor; temperature prediction; nested gated recurrent unit; convolutional neural network

分类号:

TP183; U469.72

DOI:

10.3969/j.issn.1671-8798.2022.03.003

文献标识码:

A

摘要:

为有效预测永磁同步电机的温度,首先使用嵌套结构对门控循环单元(gated recurrent unit,GRU)进行改进,提出一种嵌套式门控循环单元(nested gated recurrent unit,NGRU)网络,NGRU能对相关温度特征中的噪声进行过滤,并挖掘温度随时间变化的规律,再经过非线性变换提取深层的温度特征; 然后提出一种新型深度学习模型,即一维卷积神经网络(1D convolutional neural networks,1D-CNN)串联NGRU(1D-CNN tandem NGRU,CNGRU),CNGRU利用1D-CNN对输入特征进行初步提取,得到多角度的永磁同步电机相关温度特征作为NGRU的输入,以串联的结构融合二者的优势,得到永磁同步电机的预测温度。试验结果表明,对比其他循环网络在定子轭、定子齿和定子绕组温度上的预测表现,NGRU均方误差平均降低12.44%,无穷范数平均降低0.361 9; CNGRU在此基础上比NGRU均方误差平均降低13.29%,无穷范数平均降低0.579 6。CNGRU比NGRU及其他循环网络对永磁同步电机温度的预测,具有更高的精度和稳定性,这为保证电机的安全稳定运行提供了技术保障。

Abstract:

To efficiently predict the temperature of permanent magnet synchronous motors, firstly, the gated recurrent unit(GRU)was improved by using a nested structure to propose the nested gated recurrent unit(NGRU)network, being capable of filtering the noise in the relevant temperature features and exploring the pattern of temperature variation over time, and then extracting the deeper temperature features through a non-linear transformation; secondly, a novel deep learning model(1D-CNN tandem NGRU, CNGRU)was proposed, namely one-dimensional convolutional neural networks(1D-CNN)in tandem with NGRU. The CNGRU harnessed 1D-CNN for the initial extraction of input features to obtain multi-angle temperature features related to permanent magnet synchronous motors, serving as input to the NGRU to obtain the predicted temperature of permanent magnet synchronous motors by fusing the advantages of both in a tandem structure. The experimental results show that, compared with the prediction performance of other recurrent networks on the stator yoke, the stator teeth and the stator winding temperature, the NGRU mean square error is reduced by 12.44% on average and the infinite horizon by 0.361 9 on average; on this basis CNGRU reduces the mean square error by an average of 13.29% and the infinite horizon by an average of 0.579 6 in contrast to NGRU. It is concluded that CNGRU has higher accuracy and stability than NGRU and other recurrent networks for predicting the temperature of permanent magnet synchronous motors. It provides the technical guarantee to ensure the safe and stable operation of the motors.

参考文献
/References:

[1] 孙竟成.基于热网络法的永磁同步电机三维温度场研究[D].天津:天津大学,2016.
[2] SI J, ZHAO S, FENG H, et al. Analysis of temperature field for a surface-mounted and interior permanent magnet synchronous motor adopting magnetic-thermal coupling method[J].CES Transactions on Electrical Machines and Systems,2018,2(1):166.
[3] FANG J, DING S, SUN Y, et al. Signal injection method without torque ripple for stator winding temperature estimation of surface-mounted PMSM drive systems[J].Journal of Power Electronics,2020,20(6):1504.
[4] 丁树业,朱敏,江欣.永磁同步电机三维全域温度场与温度应力耦合研究[J].电机与控制学报,2018,22(1):53.
[5] 刘平,王鑫,孙千志,等.永磁同步电机定子绕组温度估计的信号注入策略优化[J].电机与控制学报,2019,23(11):18.
[6] 丁树业,江欣,朱敏,等.基于集总参数热网络法的永磁同步电机启动及稳态温升分析[J].电机与控制学报,2020,24(5):143.
[7] BALAMURALI A, KUNDU A, CLANDFIELD W, et al. Non-invasive parameter and loss determination in PMSM consideringthe effects of saturation, cross-saturation, time harmonics and temperature variations[J].IEEE Transactions on Magnetics,2021,57(2):340.
[8] WALLSCHEID O, KIRCHGÄSSNER W, BÖCKER J. Investigation of long short-term memory networks to temperature prediction for permanent magnet synchronous motors[C]//2017 International Joint Conference on Neural Networks(IJCNN). Anchorage:IEEE,2017:1940.
[9] HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J].Neural Computation,1997,9(8):1735.
[10] KIRCHGÄSSNER W, WALLSCHEID O,BÖCKER J. Deep residual convolutional and recurrent neural networks for temperature estimation in permanent magnet synchronous motors[C]//2019 IEEE International Electric Machines & Drives Conference(IEMDC). Santiago:IEEE,2019:1439.
[11] GUO H, DING Q, SONGY, et al. Predicting temperature of permanent magnet synchronous motor based on deep neural network[J].Energies,2020,13(18):4782.
[12] WANG X F, ZHANG Y. Multi-step-ahead time series prediction method with stacking LSTM neural network[C]//2020 3rd International Conference on Artificial Intelligence and Big Data(ICAIBD). Chengdu:IEEE,2020:51.
[13] WANG X, XU L, CHEN K. Data-driven short-term forecasting for urban road network traffic based on data processing and LSTM-RNN[J].Arabian Journal for Science and Engineering,2019,44(4):3043.
[14] CHUNG J, GULCEHRE C, CHO K H, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[EB/OL].(2014-12-19)[2021-06-10].https://arxiv.org/abs/1412.3555.
[15] KIRANYAZ S, AVCI O, ABDELJABER O, et al. 1D convolutional neural networks and applications:a survey[J].Mechanical Systems and Signal Processing,2021,151:107398.
[16] WALLSCHEID O, BÖCKER J. Global identification of a low-order lumped-parameter thermal network for permanent magnet synchronous motors[J].IEEE Transactions on Energy Conversion,2015,31(1):354.

备注/Memo:

收稿日期:2021-06-10
基金项目:教育部人文社会科学研究一般规划基金项目(17YJA880004)
通信作者:岑 岗(1959— ),男,浙江省象山人,教授,主要从事教育信息科学与技术和人工智能研究。E-mail:gcen@163.com。