摘要
针对传统剩余使用寿命预测模型难以解决长时依赖问题以及不同特征组合对模型预测精度影响大的问题,提出一种基于特征选择与Transformer的剩余使用寿命预测模型。首先利用以互信息为理论基础的最大相关最小冗余特征选择算法,捕获特征与标签、特征与特征的关系,得到最佳特征组合;然后以Transformer的编码器为主体并加入门控卷积单元组成预测模型,使模型可以充分捕捉全局信息且提高运算效率的基础上也更加注重局部信息;通过网格搜索与粒子群算法确定模型超参数。最后将最优特征组合的变量数据输入模型实现涡扇发动机剩余使用寿命预测。利用此方法在C-MAPSS数据集进行验证,并进行对比实验,结果表明预测误差与模型效率均有一定改进。
Aiming at the problem that the traditional residual service life prediction model is difficult to solve the problem of long-term dependence and that different feature combinations have a great impact on the prediction accuracy of the model,a residual service life prediction model based on feature selection and Transformer was proposed.The maximum correlation and minimum redundancy feature selection algorithm based on mutual information was used to capture the relationship between features and labels,features and features,and the best feature combination was obtained.Then taking Transformer's encoder as the main body and adding the gated convolution unit,a prediction model was formed,so that the model could fully capture the global information and improve the operational efficiency,and pay more attention to local information.The model parameters were determined by particle swarm optimization.Finally,the variable data of the optimal feature combination were input into the model to realize the prediction of the remaining service life of the turbofan engine.This method was verified in C-MAPSS data set,and comparative experiments were carried out.The results show that the prediction error and model efficiency are improved to some extent.
作者
刘耕鑫
董辛旻
张瑞博
陈阳
LIU Gengxin;DONG Xinmin;ZHANG Ruibo;CHEN Yang(Research Institute of Vibration Engineering,Zhengzhou University,Zhengzhou Henan 450001,China)
出处
《机床与液压》
北大核心
2024年第7期208-213,共6页
Machine Tool & Hydraulics
基金
国家重点研发计划项目(2016YFF0203100)
河南省重点研发项目(212102210351)。
