基于自适应深度神经网络的永磁同步直线电机定位力计算模型  

Calculation model of detent force of permanent magnet synchronous linear motor based on adaptive deep neural network

作　　者：饶章宇 刘春元 彭珍 孙浩宸 Rao Zhangyu;Liu Chunyuan;Peng Zhen;Sun Haochen(School of Information Scienceand Technology,Zhejiang Sci-Tech University,Hangzhou,Zhejiang 310018,China;College of Information Science and Engineering,Jiaxing University)

机构地区：[1]浙江理工大学信息学院,浙江杭州310018 [2]嘉兴学院信息科学与工程学院

出　　处：《计算机时代》2023年第2期1-6,10,共7页Computer Era

基　　金：浙江省自然科学基金项目一般项目(LY19E070004);嘉兴学院大学生科研训练计划(SRT)项目(8517221297)。

摘　　要：为了提高永磁同步直线电机(PMSLM)定位力计算模型的训练效率和精度,提出一种基于自适应深度神经网络(ADNN)的定位力计算模型。利用有限元参数化计算出不同结构PMSLM的定位力作为样本数据;使用一种k折训练方式结合神经网络结构搜索算法使ADNN模型结构自适应,再对ADNN模型训练得到定位力计算模型。实验结果表明,该ADNN模型精度达到99.86%;相较于人工调参,时间消耗减少85.17%;ADNN模型计算结果与样机测试结果总体一致,证明了此模型的有效性。A detent force calculation model based on adaptive deep neural network(ADNN) is proposed to improve the training efficiency and accuracy of the detent force calculation model of permanent magnet synchronous linear motor(PMSLM). The detent force of PMSLM with different structures is calculated as sample data using finite element parameterization. A k-fold training method combined with neural network structure search algorithm is used to make the ADNN model structure adaptive, and then the ADNN model is trained to obtain the detent force calculation model. The experimental results show that the accuracy of the ADNN model reaches 99.86%;the time consumption is reduced by 85.17% compared with the manual tuning of the parameters.The calculation results of the ADNN model are generally consistent with the test results of the prototype, which proves the effectiveness of this model.

关 键 词：永磁同步直线电机 定位力 深度神经网络 回归模型 结构自适应 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]