基于不平衡数据与集成学习的柴油机故障诊断研究

Research on Diesel Engine Fault Diagnosis Based on Imbalanced Data and Ensemble Learning

文中将集成学习理论与深度置信网络相结合,采用Adaboost-M2算法作为集成学习生成方法,使用麻雀搜索算法确定不同终止适应度下的不同初始化超参数分布的深度置信网络模型.将上述不同的深度置信网络模型作为基分类器进行集成学习,得到集成Ada-DBN模型.并通过实验分析了分类器数量对集成学习后的模型诊断性能的影响.结果表明:集成Ada-DBN模型不仅能够保证在平衡数据下的诊断能力,还能提高在不平衡数据下的泛化能力,是一种适用于实际柴油机故障诊断的有效方法.

Combining the ensemble learning theory with the deep confidence network,Adaboost-M2 algorithm was adopted as the ensemble learning generation method,and sparrow search algorithm was used to determine the deep confidence network models with different initial hyperparametric distributions under different termination fitness.The above-mentioned different deep confidence network models were used as base classifiers for ensemble learning,and the integrated Ada-DBN model was obtained.The influence of the number of classifiers on the model diagnosis performance after ensemble learning was analyzed through experiments.The results show that the integrated Ada-DBN model can not only ensure the diagnosis ability under balanced data,but also improve the generalization ability under unbalanced data,and it is an effective method for practical diesel engine fault diagnosis.