Study On Traffic Volume And Traffic Safety Risk Prediction Of Long Submarine Tunnel

With the improvement of the socio-economic level and the expansion of the scale of the complex urban road network,China’s urban motorisation level is increasing,the mileage of the cross-harbour tunnels is also increasing,the traffic congestion problem is becoming more serious and the traffic safety risk is increasing.Therefore,the use of recurrent neural network-based algorithms to build a traffic prediction model with accuracy,timeliness,portability and feasibility for extra-long cross-harbour tunnels as well as a tunnel traffic safety risk prediction model to avoid and reduce the impact of traffic congestion and accidents in cross-harbour tunnels is a major focus of current research.To this end,this paper uses machine learning and deep learning techniques to analyse the characteristics of the traffic information data obtained from actual measurements of the extra-long cross-harbour tunnel,and conducts in-depth research on traffic volume prediction,traffic safety impact factors and traffic safety risk prediction for the extra-long cross-harbour tunnel.The main research contents and findings are as follows:(1)The literature research presents the current status of research on traffic forecasting,traffic safety influencing factors and traffic safety risk forecasting in cross-harbour tunnels.The theoretical model definition and collection methods of traffic information data are briefly analysed,and the characteristics of the temporal distribution of traffic volumes and the temporal,spatial,morphological and vehicle distribution of traffic safety events in the tunnel are analysed,highlighting the uncertainty and periodicity of traffic information data,which are closely related to people’s travel needs and travel patterns,roadway conditions and composition characteristics.The analysis of traffic information data is uncertain,cyclical and closely related to people’s travel demand and mode,road conditions and composition.(2)The first-order difference of the time series of Jiaozhou Bay Tunnel traffic operation data is tested for smoothness and pure randomness,and based on the recurrent neural network algorithm combining the advantages of GRU and Transformer model algorithms and adding the self-attentiveness mechanism,an AGRU-Trans fusion model is proposed to predict the traffic volume of extra-long cross-harbour tunnel.(3)To study the ranking of traffic safety factors and their characteristic importance,the factor analysis method and random forest algorithm were improved,and a factor analysis-random forest analysis method was proposed to obtain the most realistic ranking of traffic safety factors and their characteristic importance,followed by the classification criteria for the degree of influence of each indicator on the severity of traffic incidents based on the calculated classification values,and the importance ranking of each traffic incident factor was obtained,and the results were analysed in terms of traffic volume,incident factor category,vehicle factor category,road condition category,and environmental factor category.(4)Based on the data obtained from the investigation of the traffic safety factors and the importance of the characteristics of the tunnel,a traffic safety risk prediction model based on the improved binomial and multinomial logistic regression algorithm was established for the extra long cross-harbour tunnel,and the corresponding linear relationships were established for the prediction of the dichotomous and trichotomous traffic safety risks.The results show that the improved logistic model is significantly better than the common logistic model in identifying traffic safety risks in tunnels,and that the traffic safety risk prediction model constructed in this paper has good prediction performance through confusion matrix,ROC curve and AUC value.