Mining And Analysis Of Railway Equipment Accident Data Based On Deep Learning

After more than 40 years of reform and opening up,China's transportation has developed rapidly.But since entering the new era,according to the document requirements of the Ministry of transport in 2019 "pushing forward the high-quality development of transportation in the new era",China's transportation industry has reached the critical moment of climbing and crossing the barriers.The development of transportation should no longer blindly expand the production scale,but conform to the characteristics of economic development and high-quality development in the new era.We should attach importance to the basic fields of transport production and create new development conditions by integrating high and new technologies.Based on this background,this paper studies the law of railway accidents.Around railway safety,China has built the monitoring system of five major professional departments of "locomotive,locomotive,electric vehicle".Each system has produced Pb scale storage of safety data.Most of all kinds of safety data,such as voice,image,report,etc.,can be represented by text description of this unstructured data.Therefore,the current field of railway system safe.The main carrier of massive information is text data,which has the characteristics of multi-source,heterogeneous and big data.Based on deep learning technology,this paper analyzes heterogeneous data of accident report.The main research contents are as follows:(1)Aiming at the problem of mixed data in railway accident text data,a named entity recognition model based on Word Embedding + Bi-LSTM + CRF is constructed.The railway accident report not only records the unstructured text sequence data,but also mixes a large number of heterogeneous key information,such as number,time,location and so on.In this paper,the deep learning technology of word vector generation is used to solve the problem that the input sequence of railway accident text is different in structure and difficult to extract uniformly from the source of data expression.After receiving the mixed data,the model constructs the text features through the bidirectional LSTM layer,and combines with the CRF technology to limit the optimization prediction results.According to the experimental analysis,the main indexes of the model reach more than 80%,which can effectively identify the named entities of railway accidents.(2)Aiming at the common problems of data skew and unbalanced category distribution in railway accidents,an intelligent text classification model of railway accidents based on Text RCNN + Focal loss is constructed.In the case of different collection conditions and classification rules,the number of derailment accidents accounts for a large proportion in railway accidents.This paper optimizes from the algorithm level,uses focus loss function instead of cross entropy loss function iteration,and improves the weight of difficult classification samples in the training process.By integrating the neural network language model of CNN and RNN,the model conforms to the characteristics of text structure.According to the comparative experimental analysis,it can achieve better classification effect on various data sets.(3)Based on the above work,the last example of this paper analyzes the description of railway equipment accidents in the United States.The data set contains nearly 130000 pieces of big data in 35 years.It realizes the application of railway accident information extraction,railway accident frequent geographical location analysis,key accident subject analysis,railway accident structured data statistical analysis,railway accident cause analysis,railway accident visualization analysis,etc.A computer-aided decision-making method for practical field workers is tried.