Abstract: According to the characteristic of fault data of high-speed train,a dynamic feature selecting algorithm is proposed to research the measured data of the running gear(referring mainly to bogie) of high-speed train. The approach combines the advantages of Fisher ratio and fuzzy entropy dynamically, which manages to evaluate features more accurately and removes the redundant features effectively to obtain superior feature subset by weighted average method. The new approach can improve classification accuracy. Experimental results for fault data of high-speed train show that the proposed approach not only improves the classification accuracies significantly,but also strengthens the stability in low speed. The overall-precise improvement is 5. 2% after extracting the optimal feature space in average compared with that of the original feature space.

Key words: feature selection, fuzzy entropy, Fisher ratio, fault classification, similarity classifier, robustness

摘要： 针对高铁故障数据的特点,以高速列车走行部(主要指转向架)常见故障的实测数据为研究对象,提出一种动态特征选取方法。通过结合Fisher 比率和模糊熵方法对其特征空间进行评估,有效去除冗余特征,利用加权平均方法选取优化的特征子集,从而实现故障分类。实验结果表明,与Fisher 比率方法、模糊熵方法相比,该方法能提高不同列车速度下高铁故障的分类准确度及低速时的分类稳定性;与原特征空间方法相比,使用该方法提取最优特征空间后各列车速度下的分类准确率平均提高了5. 2% 。

关键词: 特征选取, 模糊熵, Fisher 比率, 故障分类, 相似性分类器, 鲁棒性

CLC Number: 

• TP391