| Due to the increasing number of vehicles in urban population,traffic congestion and car accidents are also increasing,which has become a serious obstacle to the sustained growth of social economy.In order to overcome this difficulty,short-term traffic flow prediction technology,as a key part of intelligent transportation,has been developed rapidly.However,it is difficult to improve the accuracy of short-term traffic flow prediction because of the large amount of traffic trajectory data,strong time dependence and easy to be restricted by external environmental conditions.Although there are many research methods used to process trajectory data and make short-term traffic flow prediction,these researches mainly focus on the time relationship of traffic flow,and only a few research methods involve external environmental factors,and most of these research methods directly add the external environmental factors into the prediction model at the full connection layer,which cannot grasp its characteristics well.Therefore,this paper carries out in-depth research on short-term traffic flow forecasting technology from the following aspects:(1)By using big data technology,a series of data preprocessing operations such as abnormal data filtering,coordinate system conversion and track segmentation are carried out on the original track data set,and calculation is carried out according to the preprocessed data set to complete the technical research on the relevant data of traffic flow prediction.(2)Previous research on prediction models has been directly input into the prediction model in the form of individual variables without considering the internal relationship between parameters.Therefore,in order to improve the accuracy of prediction,data fusion method is adopted to integrate the two parameters of traffic flow and speed.In the data layer,LSTM residual model and weighted algorithm are used to complete experimental operations in the data layer.(3)Considering various characteristics of traffic flow data,a graph attention network traffic flow prediction model(MFI-GAT)was proposed,which integrated various influencing factors.The model would first integrate the time characteristics of track information generated by vehicles with weather and other climatic factors,and then input it into the prediction model together with the complex road network topology.The graph attention network(GATs)is used to learn the topology of the complex road network,and the gated circulation unit(AGRUs)with the introduction of attention mechanism is used to learn the traffic flow data and weather and other climatic characteristics parameters based on the time series.Then,the time characteristics of the vehicle trajectory information and the external climate parameters are integrated to improve the prediction effect. |
