摘要
为解决传统梯度下降算法训练门控循环单元网络(GRU)时收敛速度较慢的问题,提出了一种基于Adam优化的分数阶GRU神经网络的轴承故障诊断方法。首先,将分数阶微积分引入传统整数阶GRU网络,设计了基于分数阶梯度下降的GRU网络(FGRU);其次,结合Adam算法的自适应学习率和动量机制,构建了Adam分数阶GRU网络(Adam-FGRU)。实验结果表明,Adam-FGRU不仅显著提高了GRU网络的收敛速度和诊断精度,而且在不同分数阶参数条件下均表现出较高的诊断精度,避免了由于分数阶参数选择不当引起的诊断误差。
To address the slow convergence issue of the gated recurrent unit(GRU)network when trained using traditional gradient descent algorithms,this paper proposes a bearing fault diagnosis method based on a fractional order GRU neural network optimized by the Adam algorithm.Firstly,fractional calculus is introduced into the conventional integer-order GRU network to design a fractional gradient descent GRU(FGRU).Secondly,by incorporating the adaptive learning rate and momentum mechanism of the Adam algorithm,an Adam-based fractional order GRU network(Adam-FGRU)is constructed.Comparative experiments demonstrate that the Adam-FGRU network not only significantly improves the convergence speed and diagnostic accuracy of the GRU network but also maintains high diagnostic accuracy across various fractional order parameters,avoiding the diagnostic errors caused by improper parameter selection.
作者
李赛
雷欣怡
丁芝侠
张永
LI Sai;LEI Xinyi;DING Zhixia;ZHANG Yong(School of Electrical Information,Wuhan Institute of Technology,Wuhan 430205,China;Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,China)
出处
《自动化与仪表》
2025年第4期56-60,66,共6页
Automation & Instrumentation
基金
冶金自动化与检测技术教育部工程研究中心开放基金项目(MADTOF2023B01)
武汉工程大学研究生创新基金项目(CX2023552)
国家自然科学基金面上项目(62176189)。
