基于ALSTM-MHA的航空发动机寿命预测  

Remaining Using Life Prediction of Aero-Engine Based on ALSTM-MHA

作　　者：修瑞 丁建完[1] 刘笑炎 高创 XIU Rui;DING Jianwan;LIU Xiaoyan;GAO Chuang(School of Mechanical Science&Engineering,Huazhong University of Science and Technology,Wuhan Hubei 430070,China)

机构地区：[1]华中科技大学机械科学与工程学院,湖北武汉430070

出　　处：《机床与液压》2024年第12期187-192,共6页Machine Tool & Hydraulics

基　　金：国家重点研发计划(2019YFB1706501)。

摘　　要：为了提升剩余寿命预测任务的精度,提出一种基于注意力长短时记忆网络和多头自注意力机制(ALSTM-MHA)的剩余寿命预测模型,在利用数据时序性的条件下提取特征维度的重要程度以及时间维度的相关性信息。使用C-MAPSS数据集对模型进行实验验证,并与其他方法进行对比。结果表明:ALSTM-MHA模型能够有效地提取特征及时间维度上的注意力信息,与其他方法相比,它在均方根误差和非对称评价指标上分别降低了至少0.3%和20.48%,验证了模型的可行性和有效性。In order to improve the accuracy of the remaining life prediction task,a remaining life prediction model was proposed based on attention long and short term memory network and multi-headed self-attention mechanism(ALSTM-MHA),which could extract the importance of feature dimensions and correlation information of time dimensions under the condition of using data temporality.The model was experimentally validated using the C-MAPSS dataset and analyzed in comparison with other methods.The results show that the ALSTM-MHA model can effectively extract the attention information in feature and time dimensions,and the root mean square error and asymmetric evaluation indexes are reduced by at least 0.3% and 20.48%,respectively,compared with other methods,which verifies the feasibility and effectiveness of the model.

关 键 词：航空发动机 注意力长短时记忆网络 多头自注意力机制 剩余寿命预测 

分 类 号：V271.4[航空宇航科学与技术—飞行器设计]