深度学习及其在电机控制中的研究现状与展望  

Research and Prospect on Deep Learning and Its Application in Motor Control

作　　者：曹美禾 宋琳 孙军 李忠瑞 聂子玲[1] CAO Meihe;SONG Lin;SUN Jun;LI Zhongrui;NIE Ziling(National Defense Key Laboratory of Science and Technology on Vessel Integrated Power System,Naval University of Engineering,Wuhan 430033,China)

机构地区：[1]海军工程大学、舰船综合电力技术国防科技重点实验室,湖北武汉430033

出　　处：《电机与控制应用》2023年第5期1-9,共9页Electric machines & control application

基　　金：国家自然科学基金项目(52077219)。

摘　　要：状态观测器、卡尔曼滤波器等传统观测器由于应用范围的局限性,自身结构的缺陷性,会使电机控制系统的设计变复杂,观测精度降低,观测范围受限等。深度学习具有非线性拟合能力和泛化能力好以及特征提取能力强等优点。以深度学习观测器为主线,阐述了其应用到电机控制系统中的优势,以四种典型网络为例介绍了深度学习的原理和结构,列举并重点分析了深度学习在电机控制中的应用,对未来深度学习在电机观测器控制中的发展方向进行了展望。Due to the limitations of the application range and the defects of their own structures,the traditional observers such as state observers and Kalman filters may complicate the design of motor control systems,reduce the observation accuracy and limit the observation range,etc.Deep learning has the advantages of non-linear fitting capability,good generalization ability and strong feature extraction ability,etc.Deep learning observer is taken as the main line,the advantages of its application in motor control system is described.The principle and structure of deep learning with four typical networks is taken as examples,the analysis of the application of deep learning in motor control is listed and focused on,and on the future developing direction of deep learning in motor observer control is forecasted.

关 键 词：电机控制 深度学习 神经网络观测器 卷积神经网络 

分 类 号：TM301[电气工程—电机]