基于BiGRU-AE与融入加性注意力机制的双层BiLSTM模型的航空发动机RUL预测  

RUL Prediction of Aircraft Engines Based on BiGRU-AE and a Dual-Layer BiLSTM Model with Additive Attention Mechanism

作　　者：何宜霖 He Yilin(School of Economics and Management,Hebei University of Technology,Tianjin,China)

机构地区：[1]河北工业大学经济管理学院,天津

出　　处：《科学技术创新》2025年第10期215-218,共4页Scientific and Technological Innovation

摘　　要：针对预测航空发动机剩余使用寿命时预测精度较差,受噪声和冗余特征影响较大的难题,提出结合一维卷积层和双层门控循环单元的自编码器作为数据降维模型,以及融入加性注意力机制的双层双向长短期记忆网络模型作为预测模型。首先,通过降维模型提取局部特征并捕捉时序数据中的长期依赖关系;随后,搭建BiLSTM网络,并融入加性注意力机制进行预测;最后,使用NASA公开的C-MAPSS数据集进行实验验证,结果表明所提模型在降维和预测准确性上均表现出色。This study addresses low prediction accuracy and the impact of noise in predicting the remaining use原ful life(RUL)of aircraft engines by proposing a data reduction model that integrates a one-dimensional convolution原al layer with a dual-layer Gated Recurrent Unit(GRU)autoencoder,alongside a prediction model using a dual-lay原er Bidirectional Long Short-Term Memory(BiLSTM)network with an additive attention mechanism.The dimension原ality reduction model extracts local features and captures long-term dependencies in time series data.Experiments on the NASA C-MAPSS dataset show that the proposed model significantly improves both dimensionality reduction and prediction accuracy.

关 键 词：航空发动机 自编码器 长短期记忆网络 注意力机制 剩余使用寿命 

分 类 号：V231[航空宇航科学与技术—航空宇航推进理论与工程]