|
|
|
| Prediction Method of Deep Transfer Learning Performance of Permanent Magnet Synchronous Motor under the Condition of Few-Shot |
| Jin Liang1,2, Yan Yingang1,2, Yang Qingxin1, Liu Suzhen1, Zhang Chuang1 |
1. State Key Laboratory of Reliability and Intelligence of Electrical Equipment Hebei University of Technology Tianjin 300401 China;
2. Key Hebei Key Laboratory of Electromagnetic Field and Electrical Apparatus Reliability of Hebei Province Hebei University of Technology Tianjin 300401 China |
|
|
|
Abstract A permanent magnet synchronous motor (PMSM) is a typical high-efficiency and energy-saving motor. The surrogate model can replace the numerical simulation represented by the finite element method to participate in the motor optimization design, which can improve the calculation efficiency of performance analysis and optimization and reduce resource consumption. It is the basis and focus of performance analysis and optimization. Using a deep learning algorithm to establish a surrogate model can achieve fast prediction and analysis of motor performance, but its training process requires a large number of samples. Besides, the model prediction accuracy is low when the samples are insufficient, which restricts the application of the algorithm in engineering. It is necessary to consider that the label samples accumulated in the historical task are sufficient. There is relevant knowledge information between such samples and the samples of the target task. Consequently, this paper focuses on the following aspects: (1) Establish a deep learning method for transfer performance prediction; (2) Apply the performance prediction knowledge transfer accumulated in the historical motor samples to the performance analysis of the target motor; (3) Realize the performance prediction task of the target motor under the condition of a few-shot.
In electrical equipment performance prediction, there are many unlabeled feature samples and few labeled samples. Combined with the deep belief network (DBN) and the model transfer method, DBN can self-learn the characteristics of unlabeled samples in the target domain; model transfer can overcome the distribution difference between different data and solve the problem of scarcity of labeled samples in the target domain. The historical motor data are set as the source domain and the target motor data as the target domain. Firstly, the DBN is trained layer by layer unsupervised using the unlabeled samples of the target motor to obtain highly abstract important features. Secondly, a parameter-sharing pre-training network is established using the labeled samples of historical motors. Finally, the model transfer is realized by freezing the network, training the adaptation layer, and fine-tuning the whole network. Moreover, a deep transfer learning model is obtained to effectively predict the target domain data.
Taking 2004, 2010, and 2017 Prius vehicles PMSMs as the research object, using two-dimension finite element analysis (2D-FEA) as the calculation tool, the data set is obtained by analyzing and calculating the motor output power and torque ripple under different design parameters. Considering the degree of structural difference between different motors, the case of similar and differential motor groups is verified, and the generalization ability of the model is further tested. Through the comparison and analysis between several models, the progressiveness and effectiveness of the method are verified. When only 5% labeled samples of the target motor are used, the mean absolute percentage error (MAPE) of output power and torque ripple for the similar motor group are 1.54% and 2.77%, respectively. Similarly, the MAPE values of output power and torque ripple for differential motor groups are 2.73% and 4.37%, respectively. When the precision is close to that of the traditional algorithm, the target domain label samples can be reduced by 35%~55%. The results show that the proposed method can predict the performance of the target motor under the condition of a few-shot.
|
|
Received: 28 July 2022
|
|
|
|
|
|
[1] Guo Hong, Xu Jinquan, Chen Y H.Robust control of fault-tolerant permanent-magnet synchronous motor for aerospace application with guaranteed fault switch process[J]. IEEE Transactions on Industrial Elec-tronics, 2015, 62(12): 7309-7321.
[2] 李晓华, 赵容健, 田晓彤, 等. 逆变器供电对电动汽车内置式永磁同步电机振动噪声特性影响研究[J]. 电工技术学报, 2020, 35(21): 4455-4464.
Li Xiaohua, Zhao Rongjian, Tian Xiaotong, et al.Study on vibration and noise characteristics of interior permanent magnet synchronous machine for electric vehicles by inverter[J]. Transactions of China Electrotechnical Society, 2020, 35(21): 4455-4464.
[3] 曹春堂, 兰志勇, 沈凡享. 永磁同步电机无位置传感器控制系统中初始位置角检测综述[J]. 电气技术, 2020, 21(6): 1-6.
Cao Chuntang, Lan Zhiyong, Shen Fanxiang.Review of initial position angle detection in sensorless control system of permanent magnet synchronous motor[J]. Electrical Engineering, 2020, 21(6): 1-6.
[4] 谢冰川, 张岳, 徐振耀, 等. 基于代理模型的电机多学科优化关键技术综述[J]. 电工技术学报, 2022, 37(20): 5117-5143.
Xie Bingchuan, Zhang Yue, Xu Zhenyao, et al.Review on multidisciplinary optimization key tech-nology of electrical machine based on surrogate models[J]. Transactions of China Electrotechnical Society, 2022, 37(20): 5117-5143.
[5] 乔俊飞, 王功明, 李晓理, 等. 基于自适应学习率的深度信念网设计与应用[J]. 自动化学报, 2017, 43(8): 1339-1349.
Qiao Junfei, Wang Gongming, Li Xiaoli, et al.Design and application of deep belief network with adaptive learning rate[J]. Acta Automatica Sinica, 2017, 43(8): 1339-1349.
[6] 仲林林, 胡霞, 刘柯妤. 基于改进生成对抗网络的无人机电力杆塔巡检图像异常检测[J]. 电工技术学报, 2022, 37(9): 2230-2240, 2262.
Zhong Linlin, Hu Xia, Liu Keyu.Power tower anomaly detection from unmanned aerial vehicles inspection images based on improved generative adversarial network[J]. Transactions of China Elec-trotechnical Society, 2022, 37(9): 2230-2240, 2262.
[7] Yan Jiongcheng, Li Changgang, Liu Yutian.Insecu-rity early warning for large scale hybrid AC/DC grids based on decision tree and semi-supervised deep learning[J]. IEEE Transactions on Power Systems, 2021, 36(6): 5020-5031.
[8] Lu Na, Li Tengfei, Ren Xiaodong, et al.A deep learning scheme for motor imagery classification based on restricted Boltzmann machines[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2017, 25(6): 566-576.
[9] Beers A, Brown J, Chang Ken, et al.DeepNeuro: an open-source deep learning toolbox for neuroima-ging[J]. Neuroinformatics, 2021, 19(1): 127-140.
[10] 卢锦玲, 郭鲁豫. 基于改进深度残差收缩网络的电力系统暂态稳定评估[J]. 电工技术学报, 2021, 36(11): 2233-2244.
Lu Jinling, Guo Luyu.Power system transient stability assessment based on improved deep residual shrinkage network[J]. Transactions of China Elec-trotechnical Society, 2021, 36(11): 2233-2244.
[11] 金亮, 王飞, 杨庆新, 等. 永磁同步电机性能分析的典型深度学习模型与训练方法[J]. 电工技术学报, 2018, 33(增刊1): 41-48.
Jin Liang, Wang Fei, Yang Qingxin, et al.Typical deep learning model and training method for per-formance analysis of permanent magnet synchronous motor[J]. Transactions of China Electrotechnical Society, 2018, 33(S1): 41-48.
[12] 杨阳, 赵吉文, 宋俊材, 等. 基于深度神经网络模型的无铁心永磁同步直线电机结构优化研究[J]. 中国电机工程学报, 2019, 39(20): 6085-6094, 6189.
Yang Yang, Zhao Jiwen, Song Juncai, et al.Structural optimization of air-core permanent magnet syn-chronous linear motors based on deep neural network models[J]. Proceedings of the CSEE, 2019, 39(20): 6085-6094, 6189.
[13] Wang Yanbin, You Zhuhong, Li Xiao, et al.Predi-cting protein-protein interactions from protein sequences by a stacked sparse autoencoder deep neural network[J]. Molecular BioSystems, 2017, 13(7): 1336-1344.
[14] Li Chuan, Zhang Shaohui, Qin Yi, et al.A systematic review of deep transfer learning for machinery fault diagnosis[J]. Neurocomputing, 2020, 407: 121-135.
[15] 陈祝云. 基于深度迁移学习的机械设备智能诊断方法研究[D]. 广州: 华南理工大学, 2020.
[16] 陶卓. 面向少标签和无标签分类任务的迁移学习方法[D]. 北京: 北京邮电大学, 2021.
[17] Weiss K, Khoshgoftaar T M, Wang Dingding.A survey of transfer learning[J]. Journal of Big Data, 2016, 3(1): 1-40.
[18] 陈剑, 杜文娟, 王海风. 基于对抗式迁移学习的含柔性高压直流输电的风电系统次同步振荡源定位[J]. 电工技术学报, 2021, 36(22): 4703-4715.
Chen Jian, Du Wenjuan, Wang Haifeng.Location method of subsynchronous oscillation source in wind power system with VSC-HVDC based on adversarial transfer learning[J]. Transactions of China Electro-technical Society, 2021, 36(22): 4703-4715.
[19] 庄福振, 罗平, 何清, 等. 迁移学习研究进展[J]. 软件学报, 2015, 26(1): 26-39.
Zhuang Fuzhen, Luo Ping, He Qing, et al.Survey on transfer learning research[J]. Journal of Software, 2015, 26(1): 26-39.
[20] Dong Rongzhi, Dan Yabo, Li Xiang, et al.Inverse design of composite metal oxide optical materials based on deep transfer learning and global optimi-zation[J]. Computational Materials Science, 2021, 188: 110166.
[21] 杨鑫宇, 吕政, 赵珺, 等. 基于迁移学习的离心式水泵扬程性能预测[J]. 控制理论与应用, 2021, 38(5): 615-622.
Yang Xinyu, Lü Zheng, Zhao Jun, et al.Prediction of lift performance of centrifugal pump based on transfer learning[J]. Control Theory & Applications, 2021, 38(5): 615-622.
[22] 谭治学. 多源信息融合的民航发动机性能预测方法研究[D]. 哈尔滨: 哈尔滨工业大学, 2018.
[23] Xu Gaowei, Liu Min, Jiang Zhuofu, et al.Online fault diagnosis method based on transfer convolutional neural networks[J]. IEEE Transactions on Instru-mentation and Measurement, 2020, 69(2): 509-520.
[24] 康守强, 胡明武, 王玉静, 等. 基于特征迁移学习的变工况下滚动轴承故障诊断方法[J]. 中国电机工程学报, 2019, 39(3): 764-772, 955.
Kang Shouqiang, Hu Mingwu, Wang Yujing, et al.Fault diagnosis method of a rolling bearing under variable working conditions based on feature transfer learning[J]. Proceedings of the CSEE, 2019, 39(3): 764-772, 955.
[25] 王艳新, 闫静, 王建华, 等. 基于域对抗迁移卷积神经网络的小样本GIS绝缘缺陷智能诊断方法[J]. 电工技术学报, 2022, 37(9): 2150-2160.
Wang Yanxin, Yan Jing, Wang Jianhua, et al.Intelligent diagnosis for GIS with small samples using a novel adversarial transfer learning in convolutional neural network[J]. Transactions of China Electro-technical Society, 2022, 37(9): 2150-2160.
[26] Hinton G E.Training products of experts by mini-mizing contrastive divergence[J]. Neural Computation, 2002, 14(8): 1771-1800.
[27] Ayers C W.Evaluation of 2004 Toyota prius hybrid electric drive system interim report[R]. Oak Ridge National Laboratory, 2004.
[28] Burress T A, Campbell S L, Coomer C, et al.Evaluation of the 2010 Toyota prius hybrid synergy drive system[R]. Oak Ridge National Laboratory, 2011.
[29] Pan S J, Yang Qiang.A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engin-eering, 2010, 22(10): 1345-1359.
[30] Ganin Y, Lempitsky V S.Unsupervised domain adaptation by backpropagation[C]//International Con-ference on Machine Learning, Lille, France, 2015, 37: 1180-1189. |
|
|
|
2023, Vol. 38 
