为提高航空发动机剩余寿命(remaining useful life,简称RUL)预测能力,构建了一种注意力机制与长短期记忆网络(long short-term memory,简称LSTM)融合的深度学习模型.首先,分析多元高维的运行参数与RUL之间的协方差相关性,实现数据降维,优化模型权重;其次,利用运行参数的时序退化特性提高模型的回归预测效果.在NASA发动机数据集上实验的均方根误差(root mean square error,简称RMSE)范围为[4.83,13.66],与卷积神经网络(convolution neural networks,简称 CNN)、LSTM 和双向长短期记忆网络(bi-directional long short-term memory,简称Bi-LSTM)方法相比,极大地提高了预测的准确度,实现了超前预测.合并样本的方法提高了模型的泛化性,对不同类型的发动机RUL预测具有指导意义.
Prediction Model of Aero-engine Remaining Useful Life Based on Deep Learning Method
To improve the ability of prediction of remaining useful life(RUL),a deep learning model integrat-ing attention mechanism and long short-term memory(LSTM)network is constructed.By using the covariance correlation analysis between high-dimensional and multivariate characteristics of monitoring data and RUL,data dimension reduction and model weight optimization are achieved.At the same time,the time series degradation characteristics of monitoring data are used to improve the regression prediction effect of the model.The experi-mental results on NASA engine dataset show that the root mean square error(RMSE)range of the proposed model is[4.83,13.66].Compared with convolution neural networks(CNN),LSTM network and bi-directional long short-term memory(Bi-LSTM)network,the prediction accuracy is greatly improved and ad-vanced prediction is achieved.The method with combined samples improves the model generalization and has certain guiding significance to predict the RUL of different engine types.
aero-engineprediction of remaining useful life(RUL)covariance analysisattention mechanismlong-short term memory(LSTM)network
NETL
NSTL
万方数据
维普
