Evaluation And Prediction Model Research On Expressway Traffic Safety Status Based On Multi-Source Data

With the constant increase in the number of automobiles,the issue of highway traffic safety has gotten much worse recently.The Ministry of Transport has launched a national video cloud networking initiative in response to this issue,which aims to upload real-time monitoring footage of road segments to the cloud for incident identification.The usage of high-frame video over the whole road network would be a waste of transmission and computer resources,despite the fact that highway traffic accidents are typically low-probability events.Therefore,expressway operation and management units need to find a quick solution for how to upload videos with the appropriate frame rates and allocate computing resources fairly according to the actual operational state of road sections.In conclusion,this article examines the separation of traffic safety status and its connection to accident risk using data from multiple sources about expressways.This enables the prediction of the road section’s traffic safety state and the determination of the accident risk level pertinent to that road section,which serves as a foundation for the choice of the video frame rate and the allocation of computing resources in accordance with the accident risk level.The following is the paper’s primary contribution:Prior to preprocessing,the primary content of multi-source data,such as highway ETC gantry data,meteorological environment data,and historical traffic accident data,is evaluated.According to the theory of traffic flow,the ETC gantry data is transformed into various traffic flow characteristic parameters,the multi-source data is linked to the time and location of the accident as the correlation conditions,and the traffic accident including meteorological data and traffic flow characteristic data is obtained.data.The paired “case-control” technique was used to create the data set used in this study.Second,four variables that have an impact on traffic safety are examined: people,cars,roads,and environment.The highway traffic safety status index system is built at the same time,taking into account the influence of temporal features on traffic safety status and combining with the concept of index selection.The indicators are screened using a combination of the Spearman correlation analysis method and the random forest algorithm of Bayesian optimization,taking into account the effect of high-dimensional feature data on the model.The screened index data is then clustered using the fuzzy K-Prototypes clustering method.Class analysis to determine various levels of traffic safety.The Bayesian conditions-based logistic regression model assesses the risk of various states and categorizes them into high and low risk categories.The highway traffic safety state prediction model is then built based on the labeled data set and the traffic safety state acquired by fuzzy K-Prototypes clustering.To address the issue of sample imbalance in the data set,the best sampling approach is chosen by evaluating the model effects under various sampling techniques.The traffic safety status of an expressway is predicted using the Bayesian optimized random forest model,the XGBoost model,and the LightGBM model,and the prediction outcomes of the various models are contrasted.The study’s findings demonstrate that the LightGBM model,which is based on Bayesian optimization,has greater predictive power.