基于改进GRU的航空发动机寿命预测自注意力优化算法  

Improved GRU-based self-attention optimization algorithm for aero-engine remaining useful life prediction

作　　者：郭晓静 徐晓慧 郭佳豪 GUO Xiaojing;XU Xiaohui;GUO Jiahao(School of Aeronautical Engineering,Civil Aviation University of China,Tianjin 300300,China;College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学航空工程学院,天津300300 [2]中国民航大学电子信息与自动化学院,天津300300

出　　处：《航空动力学报》2024年第12期440-450,共11页Journal of Aerospace Power

摘　　要：航空发动机性能参数具有多元高维及时序性,可表征寿命退行,采用常规模型训练易导致梯度消失。因此提出一种改进门控循环单元(gated recurrent unit)的自注意力(self-attention)优化算法,分析数据源域行梯度及列间相关性,扩增寿命强相关列优化特征权重,加速模型收敛,提高预测精度。在发动机寿命预测数据集(C-MAPSS)上实验表明:该算法得到的寿命方均根误差(RMSE)落在区间[10.52,18.91],超前预测分值(score)落在区间[48.69,204.98],相比传统方法大幅降低,改善了寿命预测效果,能够为发动机寿命预测和超前维护提供有效解决方案。Multivariate,high-dimensional and time-ordered aero-engine performance parameters can characterize life regressions,which are prone to gradient disappearance using conventional model training.A self-attention optimization algorithm was proposed to improve the gated recurrent units(GRU).Row gradients of source domain and inter-column correlations were analyzed.The feature weights were optimized by augmenting the strongly correlated lifetime columns,with its aim to accelerate model convergence and improve prediction accuracy.Experiments on the engine life prediction dataset(CMAPSS)showed that the root mean square error(RMSE)of life obtained by the algorithm fell in the interval[10.52,18.91]and the over-prediction index(score)in the interval[48.69,204.98].Compared with the traditional method,the effect of life prediction was greatly reduced,and an effective solution was provided for engine life prediction and advanced maintenance.

关 键 词：剩余使用寿命 预测与健康管理 门控循环单元 自注意力机制 主成分分析 

分 类 号：V240.2[航空宇航科学与技术—飞行器设计]