改进深度信念网络的组合导航系统故障诊断方法  

Improved Deep Belief Network of Fault Diagnosis Integrated Navigation Systems

作　　者：赵善飞 张华强[1] 贾明玉 芦男 陈雨 ZHAO Shanfei;ZHANG Huaqiang;JIA Mingyu;LU Nan;CHEN Yu(Shandong University of Technology,Zibo 255049,China;Beijing Institute of Space Launch Technology,Beijing 100076,China)

机构地区：[1]山东理工大学,山东淄博255049 [2]北京航天发射技术研究所,北京100076

出　　处：《探测与控制学报》2024年第6期65-71,共7页Journal of Detection & Control

基　　金：青年科学基金项目(61803035)。

摘　　要：为提高INS/GNSS组合导航系统故障诊断的准确率与稳定性,提出一种基于改进深度信念网络的组合导航系统故障诊断方法。该方法基于状态χ2法对组合导航系统进行实时检测,将检测结果作为样本数据用于改进深度信念网络训练,利用深度信念网络提取数据的深层特征和故障分类。引入径向基函数作为模型的激活函数,提高深度信念网络面对复杂数据分布的适应能力;采用自适应矩估计算法代替传统梯度下降算法来提高故障诊断的准确率。数值仿真结果表明,该算法对故障识别的准确率达到了98%,能有效地对INS/GNSS组合导航系统的故障类型做出诊断,确保系统的平稳运行。In order to improve the accuracy and stability of INS/GNSS integrated navigation system,a method based on improved deep belief network was proposed.Method based on state chi-square test(SCST)was used for real-time detection of integrated navigation systems,and the detection results used as sample data to improve deep belief network(DBN)training.The deep belief network was used to extract deep features and fault classification from the data.Introducing radial basis functions(RBF)as the activation function of the model to improve the adaptability of deep belief networks to complex data distributions;using adaptive moment estimation(ADAM)algorithm instead of traditional gradient descent algorithm to improve the accuracy of fault diagnosis.The numerical simulation results showed that the accuracy of the algorithm in fault identification reaches 97%,which could effectively diagnose the fault types of the INS/GNSS integrated navigation system and ensured the smooth operation of the system.

关 键 词：组合导航系统 故障诊断 深度信念网络 径向基函数 自适应矩估计算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]