Research of System Fault Diagnosis Method Based on Imbalanced Data
Qingyu Zhu, Hengyu Liu, Junling Wang, Shaowei Chen, Pengfei Wen, Shengyue Wang · 2019
For the training of multiple models, the performance of single algorithm will be unstable, and an adaptive imbalance classification algorithm is proposed. The algorithm combines the area under curve (AUC) value to optimize the support vector description algorithm, the random forest algorithm and the gradient boosting decision tree algorithm respectively to generate the sub-model for classification. This paper uses the model fusion method to generate the sub-model for fusion. Finally, this paper selects the optimal sub-model to get stable classification results based on the AUC value. The algorithm in this paper is verified on the equipment running data set published by the American Prognostic and Health Management (PHM) Society in 2015. The result shows that the fault recognition rate is higher than that of the single imbalance classification algorithm. The classification effect is superior and the performance is stable.