基于文本数据挖掘的轨道电路故障智能分类

Intelligent Classification of Track Circuit Fault Based on Text Data Mining

针对轨道电路不均衡的故障历史文本数据,提出了一种针对非均衡历史文本数据挖掘的轨道电路智能分类模型。选取TF-IDF和先验LDA无监督机器学习模型对历史故障文本数据分别进行词项级和主题级故障特征提取并向量化,将提取的历史数据特征向量串行融合,得到其特征向量空间。采用SMOTE算法自动生成历史文本数据中的少数类数据,避免在机器学习过程中出现欠拟合现象。鉴于单个分类器在机器学习及智能分类的精度不高,采用投票的方式实现基分类器与集成分类器的集成学习。选择广铁集团电务段2011年的故障文本数据进行试验分析,验证该模型在历史故障数据分类的准确率和召回率等方面的优势。

According to the unbalanced fault history text data of track circuit,an intelligent classification model of track circuit for unbalanced history text data mining is presented. TF-IDF and priori LDA unsupervised machine learning model are selected to extract and quantify the fault features at word level and topic level respectively. The feature vectors of the fault history data are fused serially to obtain the feature vector space of the fault history data. SMOTE algorithm is used to generate a few classes of historical text data automatically,the phenomenon of under-fitting in the process of machine learning is avoided. In view of the low accuracy of single classifier in machine learning and intelligent classification,multi-classifier ensemble learning is implemented by voting. The fault text data of Guangzhou Railway Group Telecom Depot in 2011 are selected for experimental analysis to verify the advantages of the model in the accuracy and recall rate of historical fault data classification.