摘要
精确预测设备的剩余使用寿命能帮助厂商衡量所生产设备的质量,也有利于使用者及时更换或修理设备。截至2021年,基于相似性预测剩余寿命的研究大多使用欧式距离进行相似性的判定,容易出现维度灾难。本研究结合长短期记忆网络和自编码器提取设备的时序特征,之后使用监督局部模型进行剩余使用寿命预测,以随机森林为基础进行相似性的判定。最后,本研究使用C-MAPSS数据集验证了所提出方法的有效性,所提出方法在预测精度上要优于其余几个对比方法,并进行相关的讨论。
Accurate prediction of equipment’s remaining useful life helps manufactures measure the quality of produced equipment and helps users to replace or repair the equipment in time.Up to year 2021,most of the existing research on remaining useful life prediction has been based on similarity use Euclidean distance to judge similarity,which was prone to cause curse of dimensionality.In this paper,long short-term memory network and autoencoders were combined to extract equipment’s time series features,and then Supervised Local model was used to predict remaining useful life.Similarity was determined based on random forest.Finally,C-MAPSS dataset was used to prove the effectiveness of the proposed method,where the proposed method performed the best among other compared methods,and relevant discussions were conducted.
作者
陆晨昕
LU Chenxin(School of Economics and Management,University of Chinese Academy of Sciences,Beijing 100190)
出处
《科技促进发展》
2022年第8期1030-1038,共9页
Science & Technology for Development
关键词
剩余寿命预测
长短期记忆网络
自编码器
监督局部模型
随机森林
remaining useful life prediction
long short-term memory network
autoencoders
Supervised Local modeling methods
random forest
