Road Traffic Safety Risk Assessment And Countermeasures Based On Multi-source Data Mining Technology

In order to explore the correlation between road traffic accident risk and traffic related factors and implement an accurate prediction of road traffic accident risk.This paper selects the Jinyidong Highway in Jinhua City,Zhejiang Province as the research object.First,the road static data(flat curve radius,flat curve declination,flat curve length,slope gradient,vertical curve radius,vertical curve length,road surface friction coefficient,road intersection density),traffic dynamic data(AADT,vehicle type ratio,average speed,and vehicle type speed difference),other traffic data(parking line of sight)and accident data were collected for the integration of traffic data.Secondly,based on the integration of multi-source traffic data,the three kinds of missing data prediction model were established by Gradient Boosting Decision Tree(GBDT),Random Forest(RF)and Linear Regression(LR).Thirdly,the Apriori algorithm based on Hash tree was applied to identify the key factors affecting the road traffic accident risk.Then,road traffic safety risk was established by using Recurrent Neural Network(RNN)and Long Short-Term Memory(LSTM)to predict road traffic safety risks.Finally,association rules mining results and road traffic safety risk prediction models were used to analyze the relationship between key influencing factors and road traffic safety,and corresponding traffic safety improvement countermeasures are formulated.The results demonstrate that the GBDT model has good prediction effect,low mean square error and relatively high robustness,and completes the repair work of missing traffic data.Apriori would perform an excellent ability to realize the correlation analysis between traffic factors and traffic safety risks with the minimum confidence of 0.8,a total of 189 association rules are obtained,and the key factors which affects the change of road traffic safety risk are excavated including flat curve length,intersection density,AADT,vehicle type ratio and speed.About the traffic accident risk prediction model,the Root-Mean-Square-Error of the LSTM prediction model is 0.35,and the Root-Mean-Square-Error of the RNN prediction model is 0.47,which indicates that both LSTM and RNN can predict road traffic safety risks,but the prediction effect of LSTM is better than that of RNN.By changing the number of intersections,AADT and speed,and the LSTM model is used again to predict the traffic safety risk of the road when the intersection spacing is not more than 500 meters and the vehicle speed is 85Km/h.Achieving the lowest,and reducing traffic can also improve road traffic safety.The research results provide a theoretical and practical basis for machine learning and deep learning related algorithms in the field of missing data repair methods,association rule analysis,prediction problems and Jinyidong road traffic safety improvement measures.