Accident Delay Models Based On Topic Mining And Survival Analysis

With the increase of the traffic demand,many cities in China are rapidly advancing the construction of the subway facilities.Deeply analysis the main causes of the subway accident and predict the delay of the subway accident have become one of the focus on current traffic safety research.However,subway accident data is saved in text form and the traditional analysis method is difficult to deal with subway accident data with fast and efficient capability.On the other hand,due to the distribution and the large span of the delay,the prediction accuracy of the traditional prediction model is not high and it is unreasonable in the selection of regression variables due to lack of the description of the accident cause.Therefore,the paper combines the topic mining method with the prediction model to research the subway delay.Using the topic model to extraction of the causes of the subway accident delay and deeply analysis the main causes of the subway accident to provide accident prevention reference.Predicting the delay of the subway accident to provide decision in scientific handling with the accident.The main tasks are as follows:(1)Based on mathematical statistical method to explore the distribution rules and the characteristics of subway accident and analysis of the influencing factors on subway accident.Firstly,introduce the source of the accident data and clean the accident data.Secondly,analyze the accident distribution from the perspective of the spatiotemporal distribution and overall characteristics of the accident are analyzed from the perspective of accident level,category and cause.Finally,based on the accident data,the factors of the accident are analyzed from four aspects: line,station,passenger flow and equipment.(2)Based on topic model to analyze the cause of subway accident.Firstly,based on the shortcoming of the traditional LDA topic model to propose an accident cause topic model based on the joint distribution of subway accident date,line,station,description and delay.The maximum likelihood estimation was used to estimate the parameters and the EM algorithm was used to iteratively calculation.Secondly,put forward accident prevention suggestions based on the mining result.Finally,the topic model evaluation index was introduced.According to the experimental result,it is found that the evaluation indexes of the accident cause topic model are better than the traditional LDA topic model.When the number of the topic is less,the advantage is more significant.(3)Considering the topic of accident cause and heterogeneity to predict delay.Using the accelerated failure model to establish the delay model based on the topic model results,found the best mathematical form of the delay model,analyzing the key factors on the subway accident and influence level.Considering the heterogeneity of the influencing factors to establish the model.The result show that the accident cause topic-AFT model can reasonably predict and explain the subway accident.After considering heterogeneity,the model prediction accuracy and goodness of fit are improved...