摘要
铁路安全问题是铁路运输保证的核心问题,铁路安全问题非结构化文本数据量大,文本内容无特定规律,对于综合分析解决安全故障问题造成很高的难度。针对铁路安全问题数据智能分类,提出进化集成分类器模型。运用TF-IDF算法,通过分析接触网安全问题的数据特点提取文本特征。采用决策树作为基分类器的Bagging集成分类器将文本数据分类,在Bagging分类过程中,针对Bagging算法产生的基分类器组合解集,提出采用遗传算法(Genetic Algorithm)将其优化,产生分类结果较优的基分类器组合解集。以铁路局实际安全问题进行实验分析,实验证明TF-IDF+Bagging+Genetic Algorithm=Evolutionary Ensemble Classifier进化集成分类器模型在铁路安全问题文本分类中具有较高的准确性。
Railway safety is the core of railway transportation guarantee.The unstructured text data of railway safety problems is large,and the content of the text has no specific rules,which makes it very difficult to comprehensively analyze and solve the safety problems.Aiming at the intelligent classification of railway safety data,an evolutionary ensemble classifier model is proposed.By analyzing the characteristics of the catenary security issues of data,TF-IDF model is adopted to realize the feature extraction.Bagging ensemble classifier which uses Decision Tree as the base classifier classifies the text data,in the process of classification of Bagging,for the combined solution set of base classifier generated by Bagging Algorithm,Genetic Algorithm is proposed to optimize it to generate the combined solution set of base classifier with better classification results.Based on the safety problem of power supply contact network of a railway bureau,the experimental analysis shows that the TF-IDF+Bagging+Genetic Algorithm=Evolutionary Ensemble Classifier model has a high classification index in the text classification of railway safety problems.
作者
高凡
王富章
张铭
赵俊华
李高科
Gao Fan;Wang Fuzhang;Zhang Ming;Zhao Junhua;Li Gaoke(China Academy of Railway Science,Beijing 100081,China;Beijing Jingwei Information Technologies Co.,Ltd.,Beijing 100081,China)
出处
《电子技术应用》
2021年第4期71-76,81,共7页
Application of Electronic Technique
基金
中国铁道科学研究院集团有限公司科技研究开发计划项目(2051DZG001,2019YJ115,2019YJ125)
中国国家铁路集团有限公司科研专项课题(J2019X005)
中国铁路总公司科技研究开发计划(P2018G050)
中国铁道科学研究院集团有限公司电子计算技术研究所产品开发课题(2052DZ1201)。
关键词
铁路安全问题
TF-IDF
基分类器
集成分类器
进化集成分类器
software railway safety problems
TF-IDF
base classifier
integrated classifier
evolutionary integration classifier