杨 硕,高 成.基于长短期记忆网络与轻梯度提升机的航空发动机大修期内剩余寿命预测[J].航空发动机,2024,50(3):87-92
基于长短期记忆网络与轻梯度提升机的航空发动机大修期内剩余寿命预测
Remaining Useful Life Prediction of Aeroengine during Overhaul Based on Long Short-Term Memory Network and Light Gradient Boosting Machine
  
DOI:
中文关键词:  剩余寿命预测  组合模型  轻梯度提升机  长短期记忆网络  航空发动机
英文关键词:remaining useful life prediction  combined model  light gradient boosting machine  long short-term memory network  aeroengine
基金项目:
作者单位
杨 硕,高 成 沈阳工业大学 化工过程自动化学院辽宁辽阳 111003 
摘要点击次数: 210
全文下载次数: 174
中文摘要:
      针对航空发动机大修期内由性能主导的剩余使用寿命预测中复杂设备具有状态变量多、非线性特征严重的特点以及单 一模型面临特征提取不充分、预测精度不足等问题,提出一种长短期记忆网络(LSTM)与轻梯度提升机(LightGBM)的组合新模型 方法进行大修期内剩余使用寿命(RUL)预测。通过LSTM对原始数据进行特征提取,将LSTM的输出门中特征提取后的数据作为 LightGBM模型的输入进行RUL预测。利用NASA提供的发动机实测数据集进行了仿真试验,实现了对单个发动机的RUL预测, 并与其他6种模型预测结果进行对比,对其预测剩余使用寿命的有效性进行验证。结果表明:LSTM和LightGBM组合模型比其他 模型的预测误差显著减小,其4组数据集均方根误差仅为12.45、20.23、12.58、21.75。
英文摘要:
      In response to the challenges of predicting the Remaining Useful Life (RUL) during the overhaul period of aeroengines, characterized by numerous state variables and significant nonlinear features, and the limitations of single models in insufficient feature extraction and inadequate prediction accuracy, a novel combined model approach integrating Long Short-Term Memory networks (LSTM) and Light Gradient Boosting Machine (LightGBM) for RUL prediction was proposed. The proposed method utilizes LSTM for initial feature extraction from raw data, where the features extracted from the output gate of LSTM are subsequently fed into the LightGBM model for RUL prediction. Simulation experiments were conducted using real engine datasets provided by NASA to predict the RUL of individual engines. The efficacy of the model was validated by comparing its predictions with those of six other models. The results show that the combined LSTM and LightGBM model significantly reduces prediction errors, achieving Root Mean Square Error (RMSE) values of 12.45, 20.23, 12.58, and 21.75 across four datasets, thereby outperforming other models.
查看全文  查看/发表评论  下载PDF阅读器