| The CTC station system is one of the most important train signal systems.This system can automaticly control train approach and shunting train.It can also set temporary speed limit order and transmit related information,etc.So,it is closely related to train operation safety and efficiency.At present,the fault diagnosis of CTC station system mainly relies on human,this method is greatly affected by human factors and has low efficiency.Train delay events caused by the fault of CTC station system occur from time to time.Most of the researches on this system are devoted to summarizing the experience of experts and hardly use the intelligent diagnosis method.According to the collected real fault maintenance log,this thesis studies the fault diagnosis of CTC station system to improve the work efficiency.The specific research content is as follows:(1)Combine the characteristics of word vector and Text Rank algorithm,so the word weight is calculated from word frequency and word meaning to extract the characteristic words.Experiment shows that the improved algorithm has higher accuracy,recall rate and F value than the original algorithm.(2)Analyze the characteristic of the text based on Guangzhou-Shenzhen Line’s CTC station system fault maintenance log,and put forward a root-cause mining scheme: on improving word weight calculation results with information entropy automatically generate key words and use them clean dictionary corpus.Then this scheme automatically mark the original log’s root-cause based on the text similarity theory,finally transform text to numerical information.(3)Determine the fault diagnosis network of CTC station system based on bayes.According to the marked results,this thesis determines the research nodes of the bayesian network.First,a bayesian network based on expert knowledge was established.Then,K2 algorithm and MH algorithm were modified based on prior knowledge to obtain two different bayesian networks.Finally,fuse the three networks based on majority voting to get the final bayesian network.(4)Establish a complete bayesian fault diagnosis model.This thesis analyzes and compares the results of parameter learning by combine variable elimination and join tree with EM algorithm through experiment.It proves that variable elimination is more suitable for the model.Therefore,the posterior parameters of bayesian network are determined by EM algorithm and variable elimination,and a complete bayesian network fault diagnosis model is obtained.Finally,take the diagnostic model as core realize the CTC station system fault diagnosis tools. |
