【摘要】 针对故障特诊复杂多样和故障样本类别不平衡,使用传统故障诊断算法对机械设备进行故障预测时预测结果正确率和可靠性偏低的问题,提出一种改进的Stacking集成学习结构,一方面对数据集进行降采样来重构数据集,另一方面使用贝叶斯优化和网格搜索来调节单个机器学习模型参数,最后通过改进的Stacking集成学习框架将训练好的强机器学习模型融合在一起,实现对机械设备故障的预测。在斯堪尼亚卡车空压系统故障数据集上进行实验,获得了比LightGBM、CatBoost等任一单模型更好的效果,较适用于复杂工况下的机械设备的故障诊断,为准确率和可靠性要求高的场景提供一种解决方案,具有较强的应用价值。更多还原

【Abstract】 Aiming at the complexity and diversity of fault diagnosis and the imbalance of fault sample categories,and the low accuracy and reliability of prediction results when using traditional fault diagnosis algorithms to predict mechanical equipment faults, an improved stacking integrated learning structure is proposed. On the one hand, the data set was downsampled to reconstruct the data set; on the other hand, Bayesian optimization and grid search were used to adjust parameters of single machine learning model. Finally, through the improved Stacking ensemble learning framework, the trained strong machine learning models were fused together to realize the prediction of mechanical equipment failures. Experiments on the fault data set of Scania truck air compression system have obtained better results than any single model such as Light GBM and Cat Boost. It is more suitable for fault diagnosis of mechanical equipment under complex working conditions. It provides a solution for scenes requiring high accuracy and reliability,and has strong application value.更多还原