Research On Railway Accident Causation Based On Long Short Term Memory Network And Complex Network

The railway transportation system plays a vital role in China's transportation system,and the safe operation of the railway system has always been a major concern.With the application of more and more automation and intelligent new technologies and equipment,the railway system can store a large amount of operational data.Therefore,by studying a large number of past accident data,grasping the interaction between accident causations and the occurrence mechanism of accidents is of great significance for reducing the accident rate and ensuring the safe operation of the railway system.This paper proposes a method for analyzing the causes of railway accidents based on text mining and complex network theory.The main work of the paper is as follows:(1)Based on the Bi-LSTM-CRF model,a joint extraction model of causative events and event relationships was constructed.By treating the text segment describing the causative event as an entity,using the idea of named entity recognition to convert the event and the relationship of the events into sequence prediction tasks,and at the same time propose a new labeling method for the problem of overlapping relationship labels.The relationship extraction was transformed into a multi-label classification problem,and a joint extraction model of causative events and event relationships based on the BiLSTM-CRF model was constructed to output structured dataset.Accident text data experiments show that the constructed joint extraction model has better extraction effect.(2)Based on the causative events and the interaction relationship between the events,a causal event chain describing the process of the accident was constructed.The event descriptions with high text similarity were merged to identify the influencing factors that caused the accident.Statistics the frequency of interactions between events was established,a directional weighted railway accident causation network model was constructed and the structural characteristics of the model were analyzed.(3)Facing the characteristics of the railway accident cause network model,a RACNNR algorithm for ranking the importance of railway accident causative nodes was proposed to identify the key causes that have a greater impact on the occurrence of various accidents.Compared with the commonly used node sorting algorithm,the RACNNR algorithm not only considers the weight of the edges,but also considers the different roles between the nodes,which can more fully reflect the impact of the nodes in the network and the importance of the nodes for various types of accidents.(4)In order to explore the dynamic interaction between causations,based on the railway accident causation network model and the importance of the cause nodes,a cascading failure model was constructed to simulate the propagation process of hazards between accident causes.Considering the importance of nodes,the interaction intensity between nodes,and the propagation probability of actual data,a dangerous load distribution criterion was formulated,and the critical path of dangerous evolution propagation with the greatest impact was identified.