Traffic Volume Prediction Under Traffic Accidents Considering Traffic Impact Heterogeneity

With the rapid development of economy,traffic demand is also increasing,resulting in frequent traffic accidents,which not only cause loss of peples’ s lives and property,but also cause traffic congestion and event traffic interruption.Therefore,rapid and accuration prediction of traffic volume under accidents can provide decision-making support for traffic management department to quickly conduct traffic dispersion.At present,traffic volume prediction studies under accidents are generally divided by whether accidents occur,respectively predicting normal traffic flow and traffic volume under accidents,and all accident data are put in one data set,ignoring the heterogeneity of traffic volume variation trend of different accidents,which still has shortcomings in prediction accuracy and applicability.To solve these problems,a traffic volume prediction method considering accident classification is proposed in this paper.Firstly,the historical traffic accident data set is established according to the real 122 alarm information and bayonet data,which includes the accident characteristic information,traffic volume data under the accident and traffic accident duration.Then,through the heterogeneity analysis of the impact of traffic accident time,space,its own attributes and weather conditions on traffic flow,it is proposed to use the duration of traffic accidents to classify traffic accidents,and provides a number of methods.Based on a real dataset,traffic accident classification methods are constructed based on random forests.By comparing the classification results,the random forest model that divides accidents into four types of classification methods is selected as the accident classification model in this thesis.Then,the time series of traffic volume under historical traffic accidents were put into four data sets according to the accident categories,and four types of BP neural network prediction models were constructed by taking these four data sets as samples.Finally,through case analysis to compare the prediction results between the proposed method and the traditional method without considering the heterogeneity of traffic effects.The results indicated that the error of predicted results of the proposed method are all lower than those of the traditional prediction methods,and the prediction accuracy is improved by 29.67%.The method proposed in this article helps to predict traffic volume under accidents more quickly and accurately,providing decision-making support for traffic management departments.