Multiparameter Identification of Permanent Magnet Synchronous Linear Motors Based on Multi-Strategy Optimization Neural Network
Cao Meihe1,2, Nie Ziling2, Yang Kai1, Li Huayu2, Luo Cheng1
1. School of Electrical and Electronic Engineering Huazhong University of Science and Technology Wuhan 430074 China; 2. National Key Laboratory of Science and Technology on Electromagnetic Energy Naval University of Engineering Wuhan 430032 China
Abstract：To address online multi-parameter identification of permanent magnet synchronous linear motors (PMSLM) under external disturbances and limited excitation, this paper proposes a multi-parameter identification method based on a multi-strategy optimized neural network. The measurable inputs are the mover speed, reconstructed voltage, and stator current, and the key electrical parameters, including stator resistance, stator inductance, and permanent-magnet flux linkage, are the outputs. The goal is to enable windowed online inference and bounded-latency parameter updates without introducing additional sensors or interrupting normal operation. A brief, full-rank identification window is embedded in the conventional id=0 vector-control regime. Specifically, a short-window d-axis current reference bias is applied to provide additional excitation, improve the rank condition, and restore identifiability for the target parameters. After the window, the drive returns to id=0, preserving the original operating mode while improving the information content of measured signals. Consequently, a full-rank model is established for the target parameters, and a unified preprocessing pipeline is constructed to support sliding-window estimation aligned with industrial sampling schedules. The proposed estimator employs a multi-strategy-optimized neural network (MS-Net) tailored for PMSLM identification. The architecture introduces cascaded connections to retain physically meaningful cues and mitigate gradient attenuation across layers, while allowing the model to capture coupling patterns among voltage, current, and speed under weak excitation. Genetic-algorithm initialization is adopted to place the network into a favorable basin, reduce warm-up epochs, and improve training repeatability. Dropout regularization is further incorporated to curb overfitting across operating points and parameter drift conditions. To enhance deployment robustness, voltage reconstruction and normalization are performed consistently for both training and online inference. The estimated parameters are then post-processed via projection to feasible ranges, rate limiting, and exponential smoothing before updating the controller, thereby preventing abrupt retuning and ensuring safe online adaptation. Training data are generated using a literature-based, temperature-driven drift profile for resistance, inductance, and flux linkage, capturing gradual parameter variation during thermal evolution. Validation and experimental verification are conducted on a hardware-in-the-loop platform with an actual PMSLM. The evaluation emphasizes both steady-state accuracy and transient behavior. Compared with recursive least squares (RLS), model reference adaptive system (MRAS), and extended Kalman filter (EKF), the proposed method achieves a normalized mean absolute error of no more than 1.1%. Compared with the baseline strategies, training efficiency and stability improve by approximately 37% and 20%, respectively, and the convergence success rate increases from 14% to 78%. Moreover, the estimator maintains fast convergence and small steady-state bias under weak excitation and mild measurement mismatch. In particular, MS-Net shows higher accuracy and consistency than BP/RLS/MRAS and exhibits smaller systematic bias than EKF in estimating resistance and flux linkage. In summary, the short-window d-axis current reference bias provides a practical means to recover full-rank identifiability without compromising id=0 operation. The multi-strategy network design improves convergence behavior and reduces bias with modest computational overhead, while the windowed online implementation matches industrial sampling schedules and supports bounded-latency parameter updates. The proposed framework offers a practical approach to online parameter management for PMSLM drives and can be extended to related motor-drive systems that suffer from limited excitation and disturbance-induced degradation of identification.
曹美禾, 聂子玲, 杨凯, 李华玉, 罗成. 基于多策略优化神经网络的永磁同步直线电机多参数辨识[J]. 电工技术学报, 2026, 41(20): 6946-6957.
Cao Meihe, Nie Ziling, Yang Kai, Li Huayu, Luo Cheng. Multiparameter Identification of Permanent Magnet Synchronous Linear Motors Based on Multi-Strategy Optimization Neural Network. Transactions of China Electrotechnical Society, 2026, 41(20): 6946-6957.
[1] 鲁军勇, 柳应全. 电磁发射用直线电机及其控制技术综述[J]. 电工技术学报, 2024, 39(19): 5899-5913. Lu Junyong, Liu Yingquan.Review on linear motor for electromagnetic launch and its control tech-nology[J]. Transactions of China Electrotechnical Society, 2024, 39(19): 5899-5913. [2] 赵希梅, 李德豪, 金鸿雁. 基于负载力观测器的永磁直线同步电机自适应全阶动态滑模控制[J]. 电机与控制学报, 2024, 28(12): 53-61, 72. Zhao Ximei, Li Dehao, Jin Hongyan.Adaptive full order dynamic sliding mode control of permanent magnet linear synchronous motor based on load force observer[J]. Electric Machines and Control, 2024, 28(12): 53-61, 72. [3] Sun Xiaodong, Su Zonghao, Lei Gang, et al.Robust predictive cascaded speed and current control for PMSM drives considering parameter variations[J]. IEEE Transactions on Industrial Electronics, 2024, 71(9): 10235-10245. [4] 王立俊, 赵吉文, 于子翔, 等. 基于非线性时间延迟扰动估计的永磁同步直线电机无模型鲁棒位置跟踪控制[J]. 电工技术学报, 2024, 39(18): 5692-5704. Wang Lijun, Zhao Jiwen, Yu Zixiang, et al.Model-free robust position tracking control of permanent magnet synchronous linear motor based on nonlinear time delay disturbance estimation[J]. Transactions of China Electrotechnical Society, 2024, 39(18): 5692-5704. [5] 宋琳, 聂子玲, 孙军, 等. 基于参数辨识的永磁同步直线电机循环神经网络多维观测器[J]. 电工技术学报, 2024, 39(22): 7059-7072. Song Lin, Nie Ziling, Sun Jun, et al.Multidimensional observer of permanent magnet synchronous linear motor recurrent neural network based on parameter identification[J]. Transactions of China Electro-technical Society, 2024, 39(22): 7059-7072. [6] 谷鑫, 胡升, 史婷娜, 等. 基于神经网络的永磁同步电机多参数解耦在线辨识[J]. 电工技术学报, 2015, 30(6): 114-121. Gu Xin, Hu Sheng, Shi Tingna, et al.Multi-parameter decoupling online identification of permanent magnet synchronous motor based on neural network[J]. Transactions of China Electrotechnical Society, 2015, 30(6): 114-121. [7] 苏有成, 陈志辉. 基于电感扰动的三相横向磁通永磁电机参数辨识与估算位置偏差修正[J]. 电工技术学报, 2023, 38(12): 3165-3175. Su Youcheng, Chen Zhihui.Parameter identification and estimated position deviation correction of a three-phase transverse flux permanent magnet machine based on inductance perturbation injection[J]. Transactions of China Electrotechnical Society, 2023, 38(12): 3165-3175. [8] Brosch A, Hanke S, Wallscheid O, et al.Data-driven recursive least squares estimation for model predictive current control of permanent magnet synchronous motors[J]. IEEE Transactions on Power Electronics, 2021, 36(2): 2179-2190. [9] Yao Yu, Huang Yunkai, Peng Fei, et al.Position sensorless drive and online parameter estimation for surface-mounted PMSMs based on adaptive full-state feedback control[J]. IEEE Transactions on Power Electronics, 2020, 35(7): 7341-7355. [10] 刘栋良, 赵金洋, 董旭辉, 等. 扩展卡尔曼观测器下的永磁同步电机负载扰动抑制研究[J]. 电工技术学报, 2025, 40(8): 2488-2503. Liu Dongliang, Zhao Jinyang, Dong Xuhui, et al.Research on load disturbance suppression technology of permanent magnet synchronous motor under extended Kalman filter[J]. Transactions of China Electrotechnical Society, 2025, 40(8): 2488-2503. [11] 丁卓, 张和生, 汤昳琮, 等. 基于多尺度特征融合卷积神经网络的牵引电机转子断条故障诊断方法[J]. 电工技术学报, 2026, 41(2): 512-526. Ding Zhuo, Zhang Hesheng, Tang Yicong, et al.The fault diagnosis method of traction motor broken rotor bar based on multi-scale feature fusion convolutional neural networks[J]. Transactions of China Electro-technical Society, 2026, 41(2): 512-526. [12] 张立伟, 张鹏, 刘曰锋, 等. 基于变步长Adaline神经网络的永磁同步电机参数辨识[J]. 电工技术学报, 2018, 33(增刊2): 377-384. Zhang Liwei, Zhang Peng, Liu Yuefeng, et al.Parameter identification of permanent magnet syn-chronous motor based on variable step-size Adaline neural network[J]. Transactions of China Electro-technical Society, 2018, 33(S2): 377-384. [13] 李婕, 杨淑英, 谢震, 等. 基于有效信息迭代快速粒子群优化算法的永磁同步电机参数在线辨识[J]. 电工技术学报, 2022, 37(18): 4604-4613. Li Jie, Yang Shuying, Xie Zhen, et al.Online parameter identification of permanent magnet synchronous motor based on fast particle swarm optimization algorithm with effective information iterated[J]. Transactions of China Electrotechnical Society, 2022, 37(18): 4604-4613. [14] 边琦, 马建, 张梦寒, 等. 改进细菌觅食算法的永磁同步电机参数辨识[J]. 电机与控制学报, 2024, 28(2): 174-181. Bian Qi, Ma Jian, Zhang Menghan, et al.Parameter identification of permanent magnet synchronous motor based on modified bacterial foraging algo-rithm[J]. Electric Machines and Control, 2024, 28(2): 174-181. [15] Cao Meihe, Nie Ziling, Song Lin, et al.GABP-based multiparameter identification for permanent magnet synchronous linear motors[J]. Journal of Power Electronics, 2024, 24(9): 1428-1437. [16] 耿建平, 闫俞佰, 熊光阳, 等. 基于BP神经网络的永磁同步电机转矩观测器设计[J]. 电机与控制应用, 2020, 47(1): 78-83. Geng Jianping, Yan Yubai, Xiong Guangyang, et al.Design of torque observer based on BP neural network for permanent magnet synchronous motor[J]. Electric Machines & Control Application, 2020, 47(1): 78-83. [17] 赵鑫宇, 王丽梅. 永磁直线同步电机自适应分数阶终端滑模控制[J]. 电工技术学报, 2023, 38(20): 5434-5443. Zhao Xinyu, Wang Limei.Adaptive fractional-order terminal sliding mode control for permanent magnet linear synchronous motor[J]. Transactions of China Electrotechnical Society, 2023, 38(20): 5434-5443. [18] 朱红庆, 钟思恒, 陈祥林, 等. 基于高频电压注入的带辅助齿SPMSM转子机械角度无传感器辨识策略[J]. 电机与控制应用, 2023, 50(7): 35-42, 50. Zhu Hongqing, Zhong Siheng, Chen Xianglin, et al.Sensorless identification strategy of SPMSM rotor mechanical angle with auxiliary teeth based on high-frequency voltage injection[J]. Electric Machines & Control Application, 2023, 50(7): 35-42, 50. [19] 杜志叶, 黄子韧, 俸波, 等. 基于GA-BP神经网络的碳纤维复合芯导线压接缺陷识别方法[J]. 电工技术学报, 2026, 41(1): 315-328. Du Zhiye, Huang Ziren, Feng Bo, et al.Identification method for crimping defects of carbon fiber composite core wires based on GA-BP neural network[J]. Transactions of China Electrotechnical Society, 2026, 41(1): 315-328. [20] Rakhtala S M, Casavola A.Real-time voltage control based on a cascaded super twisting algorithm structure for DC-DC converters[J]. IEEE Transactions on Industrial Electronics, 2022, 69(1): 633-641. [21] Song Weizhang, Yang Yang, Qin Wenjing, et al.Switching state selection for model predictive control based on genetic algorithm solution in an indirect matrix converter[J]. IEEE Transactions on Trans-portation Electrification, 2022, 8(4): 4496-4508. [22] Roder M, de Rosa G H, de Albuquerque V H C, et al. Energy-based dropout in restricted Boltzmann machines: why not go random[J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2022, 6(2): 276-286. [23] Liu Kan, Zhu Z Q, Stone D A.Parameter estimation for condition monitoring of PMSM stator winding and rotor permanent magnets[J]. IEEE Transactions on Industrial Electronics, 2013, 60(12): 5902-5913.
2026, Vol. 41 
