作者：张豹, 刘琼, 吴细宝, 陈雯柏
作者单位：北京信息科技大学自动化学院，北京 100192
关键词：剩余使用寿命预测;xgboost;特征选择;涡扇发动机
摘要：
针对现有基于神经网络的剩余使用寿命预测方法存在训练时间较长的问题，提出一种基于xgboost(extreme gradient boosting, xgboost)算法的预测模型。首先，清洗历史数据，重构出涡扇发动机剩余使用寿命的完整退化轨迹数据；其次，分析各个特征与剩余使用寿命之间的相关性，依据零方差标准筛选可用特征；最后，通过xgboost算法建立剩余使用寿命预测模型，并采用网格搜索法优化模型参数。结果表明，基于xgboost算法的模型预测性能优于gbdt(gradient boosting decision tree)，其中，拟合优度(r2)提升了约5%；均方根误差(rmse)降低约6.83%；训练时间缩短近4/5。与cnn-lstm方法相比，虽然xgboost方法的预测精度略低，但训练时间较短，综合效率更高。
remaining useful lifetime prediction of turbofan engine based on ensemble learning
zhang bao, liu qiong, wu xibao, chen wenbai
college of automation, beijing information science and technology university, beijing 100192, china
abstract: to solve the problem of the existing remaining useful lifetime prediction models based on neural network have long training time, propose a prediction model based on xgboost algorithm. firstly, process the historical data and reconstruct the whole degradation trajectory data of remaining useful lifetime of turbofan engine. secondly, analyze the correlation between each feature and the remaining useful lifetime, and select the available features by zero-variance criteria. finally, establish the remaining useful lifetime prediction model by xgboost algorithm and optimize parameters by the gridsearch. the results show that the performance of the prediction model based on xgboost algorithm is better than that of gbdt. meanwhile, the goodness of fit(r2) increase by 5%, root mean square error (rmse) reduce by 6.83% and training time is about 5 times shorter than gbdt. compared with cnn-lstm method, xgboost method has a slightly lower prediction accuracy, but shorter training time and higher efficiency.
keywords: remaining useful lifetime prediction;xgboost;feature selection;turbofan engine
2022, 48(7):47-52 收稿日期: 2021-04-27;收到修改稿日期: 2021-07-10
基金项目: 北京市自然科学基金面上项目（4202026，3212005）
作者简介: 张豹（1995-），男，安徽宿州市人，硕士研究生，专业方向为机器学习、寿命预测
参考文献
[1] si x s, wang w b, hu c h, et al. remaining useful life estimation – a review on the statistical data driven approaches[j]. european journal of operational research, 2010, 213(1): .1-14
[2] 胡城豪, 胡昌华, 司小胜, 等. 基于mscnn-lstm的滚动轴承剩余寿命预测方法[j]. 中国测试, 2020, 46(9): 103-110
[3] 师璐. 飞行器关键部件可靠性建模与评估方法研究[d]. 西安: 西安理工大学, 2018.
[4] 赵新超, 吕卫民, 杨立峰. 基于故障机理与随机线性模型的电子产品寿命预测[j]. 兵工自动化, 2019, 38(7): 52-55
[5] 韩林洁, 石春鹏, 张建超. 基于一维卷积神经网络的轴承剩余寿命预测[j]. 制造业自动化, 2020, 42(3): 10-13
[6] 蔡忠义, 王泽洲, 张晓丰, 等. 隐含非线性退化设备的剩余寿命在线预测方法[j]. 系统工程与电子技术, 2020, 42(6): 1410-1416
[7] 司小胜, 胡昌华, 周东华. 带测量误差的非线性退化过程建模与剩余寿命估计[j]. 自动化学报, 2013, 39(5): 530-541
[8] 朱霖, 宁芊, 雷印杰, 等. 基于遗传算法选优的集成手段与时序卷积网络的涡扇发动机剩余寿命预测[j]. 计算机应用, 2020, 40(12): 3534-3540
[9] 宋亚, 夏唐斌, 郑宇, 等. 基于autoencoder-blstm的涡扇发动机剩余寿命预测[j]. 计算机集成制造系统, 2019, 25(7): 1611-1619
[10] 徐硕, 侯贵生. 基于vae-d2gan的涡扇发动机剩余使用寿命预测[j]. 计算机集成制造系统, 2022, 28(2): 417-424
[11] freund y. boosting a weak learning algorithm by majority[j]. information & computation, 1997, 121(2): 256-285
[12] 于玲, 吴铁军. 集成学习: boosting算法综述[j]. 模式识别与人工智能, 2004(1): 54-61
[13] friedman j h. greedy function approximation: a gradient boosting machine[j]. annals of statistics, 2001, 29(5): 1189-1232
[14] chen t q, guestrin c. xgboost: a scalable tree boosting system[c]//2016 acm sigkdd international conference on knowledge discovery and data mining. acm, 2016: 785-794.
