基于自动机器学习的不平衡故障诊断方法

New automated machine learning based imbalanced learning method for fault diagnosis

为了提升故障诊断模型在数据不平衡场景下的性能,提出一种基于贝叶斯优化的自动不平衡故障诊断方法。首先,构建了一种分层多模型的参数空间,探索重采样和分类器的算法组合选择和超参数优化;然后,使用基于树形结构Parzen估计器(TPE)的贝叶斯优化器进行模型的训练与优化,得到参数空间中最优的算法组合和超参数配置;最后使用最优配置模型在测试集上进行结果评估。将所提方法应用于UCI(university of California Irvine)不平衡标准数据集和滚动轴承数据集。实验通过设置多个不平衡比,对优化后的模型分类效果进行检验,并与传统的随机搜索方法进行对比。结果表明,所提方法更好地提升了模型在不平衡故障数据上的分类能力,且优化过程更加高效。

To improve the performance of fault diagnosis models in imbalance dataset,an automatic imbalance fault diagnosis method based on Bayesian optimization was proposed.A hierarchical multi-model configuration space was constructed to explored the combination selection of resampling and classifier with their hyperparameters in this configuration space.Then a Bayesian optimizer based on Tree-structured Parzen Estimator(TPE)was used to optimize model training procedure.After training,an optimal model in the configuration space was obtained.The optimal configuration model was used to evaluate the results on the test dataset.The proposed method was applied to University of California Irvine(UCI)imbalance dataset and rolling bearing dataset.Experiments evaluated classification improvement after optimization by setting multiple imbalance ratios.Comparison with random search method was also conducted.The results showed that the proposed method improved the model classification performance better in imbalance fault diagnosis dataset,and optimization process is more efficient.