摘要
姿态控制系统是卫星系统中重要的组成部分,由于其高昂的造价,发生故障会引发恶劣的影响。随着航天科技的发展,卫星姿态控制系统也逐渐复杂,其可能发生故障的概率也随之增大。针对传统神经网络故障诊断结果缺少置信度、鲁棒性较差以及易发生过拟合的缺点,在对贝叶斯统计和深度学习理论研究的基础上,提出了一种基于贝叶斯线性层与贝叶斯卷积层的Bayesian Le Net结合的网络模型。通过对卫星姿态控制系统飞轮部件的故障数据分析和处理,进而采用该模型对故障仿真,并与贝叶斯全连接神经网络与传统Le Net进行对比,实验结果表明:在飞轮可能发生的三种故障前提下,上述网络模型准确率较高,过拟合现象较轻。验证了上述网络模型的有效性。
The attitude control system is an important component of satellite systems,and due to its high cost,malfunctions can cause adverse effects.With the development of aerospace science and technology,the satellite attitude control system has become more and more complex,and the probability of its failure has also increased.Considering the lacking of confidence,poor robustness and easy overfitting in fault diagnosis of traditional neural networks,this paper proposes a new Bayesian LeNet network model which composed of a Bayesian linear layers and a Bayesian convolutional layers.After analyzing and processing of the fault data of the flywheel components of the satellite attitude control system,the Bayesian network model was used to conduct the fault simulation experiment and compared with the Bayesian fully connected neural network and the traditional LeNet.The results show that under the premise of three possible failures of the flywheel,the Bayesian neural network has better accuracy and less overfiting.The Bayesian neural network isverified.
作者
蒋强
刘恩雨
何旭
张伟
JIANG Qiang;LIU En-yu;HE Xu;ZHANG Wei(ShenyangLigong University,Shenyang Liaoning 110168,China;Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang Liaoning 110016,China)
出处
《计算机仿真》
2024年第1期64-68,共5页
Computer Simulation
基金
面向地空无人平台测控链路混沌隐蔽通信技术研究(LG202014)。
关键词
卫星姿态控制系统
故障诊断
贝叶斯神经网络
深度学习
Satellite attitude control system
Fault diagnosis
Bayesian neural network
Deep learning
