Research On Classification And Recognition Of Railway Accident Causes And Association Rules Recognition Based On Text Mining

The railway transportation system plays an important backbone role in China’s comprehensive transportation system,and the operational safety of the railway system is of paramount importance.Carry out accident learning and analysis work on historical accident data,and obtain accurate accident cause mechanism and key causes of accidents,which can prevent the recurrence of similar accidents.According to the characteristics of railway accident data,this paper conducted qualitative and quantitative analysis of accident causes based on text mining technology,designed the method of accident cause extraction and association rule mining,and determined the propagation mechanism and key causes of accidents.The main research contents are as follows:(1)Aiming at the characteristics of railway accident reports,the LDA topic model based on word weighting was designed.The model used Gaussian function to weight the low-frequency words,medium-frequency words and high-frequency words of the accident text.At the same time,it combined the perplexity and the topic variance to determine the number of accident-causing topics.The simulation results showed that the word-weighted LDA topic model proposed in this paper can accurately identify the human,technology,management,and environmental factors in railway accidents.(2)Based on the extracted accident topics and its related features,the cause of the accident was determined.On this basis,in view of the shortcomings of the current convolutional neural network used in the classification of the cause of accidents,a multilabel classification model of railway accidents based on M-CNN was constructed.The model considered the characteristics of railway-caused texts,designed the input of text feature in word-level,character-level and topic-level network three-channel,and used KMax-Pooling technology to improve the classification performance of the model.The simulation results showed that the text classification model based on M-CNN can accurately identify the cause labels of accident samples,and the classification accuracy was increased by 5% compared with traditional CNN algorithms.(3)Aiming at the problem that the traditional association rule algorithm does not consider negative association rules and accident level weights,an improved interestingness degree C＿Inter was proposed,which weighted the accident-causing factors involved in different levels of accidents,and a weighted association rule mining algorithm based on C＿Inter was constructed.Using this algorithm to analyze the association rules of the constructed structured accident data set,the positive and negative strong association rules set for the cause of railway accidents were obtained,and then the improvement measures to deal with the association rules of different causes were given.Compared with the traditional Apriori algorithm,the algorithm proposed in this paper can identify more and more meaningful strong association rules.(4)The quantitative analysis method of causal weight based on association rules and network analysis method was designed.The weighted support degree of the mined association rules was used to determine the value weight of the accident cause,combined with the weighted confidence degree and the weighted interestingness degree to determine the impact weight of the accident cause,so as to determine the comprehensive weight of the accident cause.By analyzing and calculating the weights of the causes involved in the railway accident report in this article,the accuracy and effectiveness of the method for determining the comprehensive weights of the causes proposed in this article was fully verified.Finally,through the analysis of the key causes of railway accidents,targeted improvement measures were proposed.