基于LSTM的滑动轴承实时温度预测模型  

Real-Time Temperature Prediction Model of Sliding Bearings Based on LSTM

作　　者：刘子江 王建梅[1] 宁可[1] 赵志宇 LIU Zijiang;WANG Jianmei;NING Ke;ZHAO Zhiyu(Engineering Research Center of Heavy Machinery Ministry of Education,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学重型机械教育部工程研究中心,太原030024

出　　处：《轴承》2024年第12期66-73,共8页Bearing

基　　金：国家自然科学基金资助项目(51875382)。

摘　　要：温度是影响滑动轴承正常运转的重要因素,实时监测轴承温度对测试至关重要,但实时采集轴承温度的数据量大,为应对测试或监测过程中温度传感器故障导致数据缺失的问题,提出了一种长短时记忆(LSTM)网络预测模型,并与传统BP神经网络预测模型进行对比分析。通过处理已有的温度数据预测后续的温度变化趋势,比较了2种模型的均方根误差来描述预测精度。通过实际值与预测值的相互验证,验证了LSTM网络预测模型具有良好的预测精度,其均方根误差达到0.0805,有效弥补了BP神经网络预测模型精度的不足,解决了BP神经网络样本依赖性问题。Temperature is an important factor affecting the normal operation of sliding bearings.Real-time monitoring of bearing temperature is crucial for testing,but the bearing temperature data collected in real time is large.In order to solve the problem of data loss caused by temperature sensor fault during testing or monitoring process,a long short-term memory(LSTM)network prediction model is proposed and compared with traditional BP neural network prediction model.The temperature trend at subsequent moments is predicted by processing the existing temperature data,and the root mean square error(RMAE)of two models is compared to describe the prediction accuracy.The mutual validation between actual values and predicted values verifies that the LSTM network prediction model has good prediction accuracy with RMSE as 0.0805,compensating for shortcomings of BP neural network prediction model accuracy effectively and solving the problem of sample dependence in BP neural network.

关 键 词：滑动轴承 油膜轴承 神经网络 时间序列 温度 预测 

分 类 号：TH133.31[机械工程—机械制造及自动化]