郭朝有, 许喆, 姚乾. 基于Stacking的机舱设备剩余寿命预测方法[J]. 中国舰船研究, 2022, 17(6): 118–125. doi: 10.19693/j.issn.1673-3185.02672
引用本文: 郭朝有, 许喆, 姚乾. 基于Stacking的机舱设备剩余寿命预测方法[J]. 中国舰船研究, 2022, 17(6): 118–125. doi: 10.19693/j.issn.1673-3185.02672
GUO C Y, XU Z, YAO Q. Stacking-based method for predicting remaining useful life of engine room equipment[J]. Chinese Journal of Ship Research, 2022, 17(6): 118–125. doi: 10.19693/j.issn.1673-3185.02672
Citation: GUO C Y, XU Z, YAO Q. Stacking-based method for predicting remaining useful life of engine room equipment[J]. Chinese Journal of Ship Research, 2022, 17(6): 118–125. doi: 10.19693/j.issn.1673-3185.02672

基于Stacking的机舱设备剩余寿命预测方法

Stacking-based method for predicting remaining useful life of engine room equipment

  • 摘要:
      目的  在参考中国船级社《智能船舶规范》中智能机舱定义和要求的基础上,探索机舱设备故障预测与健康管理相关技术,开展轴承剩余寿命预测方法研究。
      方法  针对常规数据驱动的轴承剩余寿命预测存在预测精度不佳的问题,利用集成学习Stacking融合策略,优选极限梯度提升(XGBoost)与人工神经网络(ANN )为基学习器,岭回归(ridge regression)为元学习器,构建R-A-X(Ridge-ANN-XGBoost)融合预测模型。选用IEEE PHM 2012 Prognostic Challenge同工况下的全寿命周期数据作为数据集设计预测性能对比实验,以MAE和R2为性能评价指标,对比研究基于单一算法、简单平均融合方式以及R-A-X融合方法的轴承剩余寿命预测性能。
      结果  结果表明,基于Stacking构建的R-A-X融合预测模型的绝对平均误差(MAE)与决定系数(R2)评价值均优于文中涉及的其他方法,预测精度最高可提升20%。
      结论  所提出的方法可提升轴承剩余寿命预测精度,对智能机舱中设备健康管理的实现具有一定的参考价值。

     

    Abstract:
      Objectives  With reference to the definitions and requirements of intelligent engine rooms in the China Classification Society Rules for Intelligent Ships, this paper studies methods for predicting the remaining useful life (RUL) of bearings in order to explore prognostic and health management technologies.
      Methods  Addressing the poor prediction accuracy of conventional data-driven methods, this study uses the Stacking fusion strategy in integrated learning to construct an R-A-X (Ridge-ANN-XGBoost, with XGBoost and ANN as the base learner, and ridge regression as the meta learner) fusion model. It then designs a prediction performance comparison experiment using the life cycle data in the IEEE PHM 2012 Prognostic Challenge under the same working conditions, with MAE and R2 used as performance evaluation indicators to compare the R-A-X fusion model with the single algorithm and average.
      Results  The results show that the prediction performance of the R-A-X fusion model are better than those of the other methods involved in this article, with an improvement effect reaching up to 20%.
      Conclusions  The proposed method can improve the accuracy of bearing RUL prediction and has certain reference value for the realization of the equipment health management of intelligent engine rooms.

     

/

返回文章
返回