Research And Application Of Regional Road Network Short-term Traffic Flow Forecast

Traffic flow prediction is an enduring research topic.The research goals of traditional traffic flow prediction methods are one or some roads,which aren't much help to the traffic flow prediction of large-scale city-level road networks.Some of them was looking for changes in traffic flow in time series,the others explored similar relationships between roads from spatial locations,which means they have failed to integrate the temporal and spatial features.Recent years,deep learning has provided a new direction.Based on it,this thesis studies the traffic flow prediction method of the regional road network and builds a regional road network traffic flow prediction system.The main work is as follows:(1)A method is proposed to mine the interrelationships between the traffic flow of each link using multi-order adjacency matrix and graph convolutional neural network.This method can calculate the traffic flow change relationship of each road and its adjacent roads in parallel,so it can support the task of traffic flow prediction for large-scale road networks;by stacking multiple graph convolution layers,a stronger model prediction ability can be obtained ability.(2)It is proposed to use the sequence-to-sequence model to mine the temporal characteristics of traffic flow on each road and predict the traffic flow.Use the attention mechanism to capture the temporal heterogeneity of traffic flow patterns and further enhance the accuracy of multi-step predictions.The calculation at each prediction time step is performed in parallel in a matrix manner,so it can also support traffic flow prediction tasks for large-scale road networks.The open source vehicle trajectory data set was used to test the model in this thesis.The experimental results show that the model performs better than comparsion models on multiple indicators.The actual data and prediction of traffic flow on multiple roads data fitting images also show that the model in this thesis has higher accuracy.(3)The application of traffic flow prediction information was studied,and a set of short-term traffic flow prediction system for regional road network was designed and implemented.The system can parse the road network structure from a third-party map file,which is convenient for the user to select the prediction area.It provides offline and real-time data upload methods.It provides the implementation of the prediction model and classic prediction model ARIMA and LSTM in this article.The user can choose the model category,Customizing some parameters trainning and using the model.Finally,the effectiveness and stability of the system are verified by system tests.