Research On Urban Road Congestion Prediction Method Based On Vehicle Trajectory Data

Making accurate road congestion forecasting can not only enable travelers to change their travel routes in advance,but also facilitate traffic managers to timely adjust traffic control measures and reduce traffic congestion.Nowadays,however,the forecasting accuracy of traffic flow forecasting model is low,and the traffic condition classification standard is unreasonable,which makes it difficult to forecast the traffic jam accurately.According to vehicle trajectory data,this paper proposes a congestion classification model based on clustering algorithm and a combination forecasting model of Average Travel Speed with second-order spatial and temporal correlation.In this way,the traffic jam level can be accurately forecasted.And in this paper,car-hailing trajectory data is used as the data source of traffic congestion prediction to break the limitations of traditional traffic data sources and reduce the inaccuracy.Firstly,the trajectory data of online car-hailing is preprocessed.In this paper,the Open Street Map is used to obtain the electronic road network data,and the road matching algorithm based on geometric projection is used to complete the road matching.In addition,the Average Travel Speed is used as the traffic congestion evaluation index.Through the position-time interpolation model,the Average Travel Speed of road section can be accurately estimated.Meanwhile,the congestion classification model is constructed based on K-means clustering algorithm.By using the estimated data obtained from the average travel speed,the congestion classification standards of different levels of roads can be obtained.Thus,the traffic jam level can be identified comprehensively and accurately.Then,in order to make up for the shortcoming that the traditional forecasting model only considers the first-order spatio-temporal correlation of traffic flow,this paper extracted the spatio-temporal characteristics of the model considering the second-order spatio-temporal correlation.In addition,on the basis of CNN model and LSTM model,a combination forecasting model of Average Travel Speed of road sections is constructed.And the dynamic weight is introduced into the combination forecasting model so that the influence of the sub-model on the final forecasting results can be adjusted in real time.Finally,in order to verify the accuracy and superiority of the model,a test set is used to verify the prediction effect of the model.By comparing the model in this paper with the firstorder combination forecasting model and the three commonly used forecasting models respectively,the following conclusions can be drawn:(1)The combination forecasting model of the Average Travel Speed of road sections proposed in this paper has a high forecasting accuracy.Its average absolute percentage error on the test set is less than 10%,and the error is lower than that of other traditional forecasting models.So it can predict the average travel speed of road sections more accurately.(2)After introducing second-order spatial features into the model features,the average absolute percentage error of the model on the two test sets was reduced by 2.97% and 3.31%respectively.It can be proved that introducing second-order spatial features into the model features can effectively improve the forecasting accuracy of Average Travel Speed of road sections.(3)By combining the forecasting results of the combined forecasting model of Average Travel Speed of road sections with the congestion classification standard,the forecasting of road congestion level can be realized,and the forecasting accuracy of traffic status on the two test sets is 83.3% and 90.0% respectively.It can be proved that the model in this paper can forecast the traffic state of road section more accurately and can meet the requirements of short-term traffic state forecasting.