Modeling And Forecasting The Impact Of Traffic Accidents On Highway Network

Since traffic safety is related to the normal operation of the road network,it has always been valued by traffic managers and participants.When an accident occurs,it not only causes loss of life and property to the participants of the accident,but also reduces the traffic capacity of the accident section,which in turn causes greater losses.By predicting the temporal and spatial impacts caused by accidents in time,traffic managers and other traffic participants can simultaneously improve the understanding of traffic conditions,and take corresponding adjustment measures in time,thereby inhibiting the possibility of greater losses caused by accidents.Therefore,it is very important to predict the impact of traffic accidents.First of all,the analyzing of the references of the scope of the accident’s impact.This article sorts out and integrates the existing research,and summarizes the advantages and disadvantages of the existing prediction models.Because the prediction model will face complex reality in the actual application,the designed prediction model is required not only to be accurate,but also to be adaptive.Based on the above content,this article chooses to build an adaptive accident impact scope prediction model based on the KNN model.Subsequently,the field collection of accident data sets.The accident data set is preprocessed,and the distribution of various influencing factors of the accident and the temporal and spatial probability distribution of accident impact are obtained by analysis.The former can guide accident prevention and on-site disposal;the latter obtains the distribution of duration and length,which conform to the log-Laplace and half-Generalized normal distributions respectively.They are all prepared for the construction of models.Then,the establishment of an adaptive forecasting model for the impact of accidents.By analyzing the overall process of the prediction model,adaptive design points are obtained.At the same time,by building a basic KNN and using it fot prediction,the room for improvement in KNN is looked for.Two kinds of weight,influencing factor weights and forecast weights,are targetedly preoposed to transform it.The influencing factor weight can be calculated by means of decision tree models(DT)and cluster analysis(CA)models;the prediction weight can be calculated by means of distance(Dist)and probability(Prob)distribution.Through the pairwise combination of influencing factor weight and prediction weight,4 new prediction models can be obtained,namely DT-Dist-KNN,DT-Prob-KNN,CA-Dist-KNN and CA-Prob-KNN.After the models’ building,in order to evaluate the prediction performance of each model,the mean absolute error(MAE),mean absolute percentage error(MAPE)and root mean square error(RMSE)are selected as evaluation indicatorsFinally,case study.By designing two experimental schemes and using the actual data,the schemes not only successfully verified the accuracy and adaptability of the designed model,but also simulated the actual performance of the designed model.The optimal prediction model in the case is DT-Dist-KNN,and its errors in predicting the duration and the spatial length of the accident’s impact are stable to less than 5min and 20 m,respectively.Compared with the basic KNN,its prediction performance improvement in duration and length reached 37.20% and 90.11%,respectively.