为了解决涡扇发动机的监测数据维数高、时间跨度长、给预测发动机剩余使用寿命带来困难的问题。本文提出了一种基于集成神经网络模型的发动机寿命预测系统，采用集成学习中的Stacking方法对单一的学习器进行集成来预测涡扇发动机的剩余使用寿命（RUL）。模型在NASA公共数据集C-MAPSS(Commercial Modular AeroPropulsion System Simulation)上进行了发动机寿命预测实验验证，并与常用的机器学习方法和单一神经网络进行了比较。实验结果表明：模型在多种评价方法上综合表现最佳，且在超前预测上表现良好。

In order to solve the difficulties in predicting the remaining useful life (RUL)of turbofan engine caused by the high dimension and long time span of the monitoring data of the turbofan engine,this paper proposes an engine life prediction system based on the integrated neural network model,which uses the stacking method in ensemble learning to integrate the single learner to predict the RUL of turbofan engine.The model is validated by an experimental test of engine useful life prediction on the NASA public data set Commercial Modular AeroPropulsion System Simulation(CMAPSS),compared with common machine learning method and single neural network.The experimental results show that the model has optimum performance in a variety of evaluation methods and does well in prediction ahead of time.

V235

国家科技重大专项（2017-Ⅰ-0007-0008)