基于Stacking集成学习的剩余使用寿命预测

Remaining useful life prediction based on stacking ensemble learning

剩余使用寿命(RUL)预测对于设备维护策略的制定有着关键作用。面对可变环境和多样的操作条件,单一寿命预测模型的性能波动较大,泛化能力弱。针对这一问题,提出一种融合多个相异模型的Stacking集成模型,纠正单一模型的预测误差。首先,对状态监测数据进行滑动时间窗口处理,获得具有时间序列信息的性能退化数据;然后,以提高模型的准确性和多样性为目标,确定基学习器的种类;最后,将梯度提升决策树(GBDT)作为元学习器,整合基学习器的预测结果,输出最终结果。基于NASA C-MAPSS数据集,对提出的集成模型进行验证,结果表明:Stacking集成模型的预测精度显著高于基学习器,与其他传统预测模型相比,也具有明显优势。

Remaining Useful Life(RUL)prediction plays a key role in the formulation of equipment maintenance strategies.In the face of variable environment and diverse operating conditions,the performance of a single life prediction model fluctuates greatly,and the generalization ability is weak.Aiming at this problem,a Stacking ensemble model integrating multiple dissimilar models was proposed to correct the prediction error of a single model.The state monitoring data was processed by sliding time window to obtain performance degradation data with time series information;with the goal of improving the accuracy and diversity of the model,the types of base learners were determined;the Gradient Boosting Decision Tree(GBDT)was used as a meta-learner to integrate the prediction results of the base learner and output the final result.Based on the NASA C-MAPSS dataset,the proposed ensemble model was verified,and the results showed that the prediction accuracy of the Stacking ensemble model was significantly higher than that of base learners,and it also had obvious advantages compared with other traditional models.