Research On Traffic Flow Prediction Based On Online Vehicle Trajectory Data

In urban life,traffic is an important issue that affects the livelihood of the community.Traffic congestion is particularly prominent,especially during peak commuting periods,when serious congestion not only affects normal production and life,but also affects citizens’ sense of security,access and well-being.The proposed intelligent transportation system provides effective management ideas to alleviate the traffic congestion problem,and the effective prediction of traffic flow can provide data reference for the core technology of the system,so the main research content of this paper is the prediction of traffic flow.Traffic flow prediction affects the effectiveness of prediction in both data acquisition and prediction model.Firstly,in terms of data acquisition,the improved Viterbi algorithm is used for map matching to obtain traffic flow parameter data.The acquisition method is based on the trajectory dataset of the net car,and after data pre-processing,map matching,manual selection of target road sections,average speed calculation,traffic flow acquisition and other operations,the traffic flow data of any road section in the map road network can be obtained.At the same time the method has the advantages of low cost,high accuracy and intelligence.Secondly,in terms of prediction,the average traffic speed of the acquired road section is firstly predicted,and a long and short time recurrent neural network(LSTM)is designed and built.Compared with the general recurrent neural network,the LSTM can effectively avoid the problem of gradient disappearance or gradient explosion during training.By predicting the speed of vehicles,the traffic situation of the target road section can be effectively predicted in the future period.In order to excel in the prediction task,the traffic flow needs to be predicted from multiple dimensions.Therefore,in addition to speed prediction,this paper also adds the task of road section traffic flow prediction,introduces convolutional neural network model(CNN),and constructs a combined CNN+LSTM prediction model based on deep learning,by constructing a data sample matrix as model input,the CNN layer extracts the temporal features of the data,and the LSTM layer is used for traffic flow prediction.The LSTM layer is used for traffic data prediction.A comparison experiment between the individual model and the combined model is also designed to verify the effectiveness of the combined model.In order to verify the feasibility and accuracy of the two prediction models,this paper uses the publicly available trajectory dataset of DDT to conduct experiments on the prediction models.The experiments show that the performance index of the LSTM model in this paper meets the requirements and has good prediction effect.By comparing the experiments,the combined model has more accurate prediction results,better performance metrics and better fit compared with the convolutional model alone and the long and short time recurrent neural network model.