基于RF-DGRU-SA的涡扇发动机剩余寿命预测  

Remaining Useful Life Prediction of Turbofan Engine Based on RF-DGRU-SA

作　　者：马鸣风 王力[2] MA Mingfeng;WANG Li(College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China;Vocational Technical Institute,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学电子信息与自动化学院,天津300300 [2]中国民航大学职业技术学院,天津300300

出　　处：《机床与液压》2023年第1期196-201,共6页Machine Tool & Hydraulics

基　　金：国家自然科学基金委员会与中国民用航空局联合资助项目(U1733119)。

摘　　要：针对涡扇发动机在退化过程中机制复杂、状态检测数据维数过高等问题,提出一种随机森林(RF)和基于自注意力机制(SA)深度门控循环单元(DGRU)相融合的涡扇发动机剩余使用寿命(RUL)预测方法。利用RF算法确定重要度阈值实现特征筛选。将筛选出的特征输入DGRU-SA模块,通过多层GRU神经网络挖掘出相关特征与目标值之间的隐藏信息,利用SA神经网络为隐藏信息添加不同大小的权重。最后,利用全连接层输出预测结果,采用CMAPSS数据集进行实验验证。结果证明:与传统的多层感知机、卷积神经网络、长短时记忆等方法相比,所提出的融合模型误差更小,具有良好的预测精度和稳定性。In order to solve the problems of complex mechanism and high dimension of state detection data in turbofan engine degradation process,a prediction method of remaining useful life(RUL)of turbofan engine was proposed based on the combination of random forest(RF)and self-attention(SA)deep gated recurrent units(DGRU).RF algorithm was used to determine the importance threshold value to achieve feature screening.The selected features were input into DGRU-SA module,and the hidden information between the relevant features and the target value was mined through the multi-layer GRU,and the SA was used to add different weights to the hidden information.Finally,the full connection layer was used to output the prediction results.Experimental verification was carried out with CMAPSS data set.The results show that the proposed fusion model has less error and good prediction accuracy and stability compared with the traditional multi-layer perceptron,convolutional neural network,long short-term memory.

关 键 词：涡扇发动机 剩余使用寿命 随机森林 深度门控循环单元 自注意力机制 

分 类 号：V271.4[航空宇航科学与技术—飞行器设计]