Research On Travel Time Estimation Method Based On Trajectory Data

Travel Time Estimation(TTE)is to estimate the time required for the trip given the relevant travel information.It is a fundamental task in intelligent transportation systems,which can reflect the current traffic conditions to a certain extent and help in traffic scheduling and decision making,and it has important research significance.The massive amount of GPS trajectory data obtained through various devices and technologies provides a solid data foundation for TTE,so this thesis studies the travel time estimation task based on trajectory data.Travel time is influenced by various factors,such as the chosen route,departure time,traffic conditions,and external factors such as weather.To more accurate estimation results,this thesis first collects multi-source heterogeneous data,such as trajectory data,urban road network data,and external factors data,and uses different preprocessing methods for different types of data to facilitate further analysis and modeling.Secondly,this thesis proposes two travel time estimation methods based on the preprocessed data combined with deep learning and meta-learning,which are noted as Travel Time Estimation Method based on Meta learning(Meta TTE)and Travel Time Estimation Method based on Multi-head Graph Attention Network(MGTTE).Meta TTE explores the grid-based traffic feature extraction and encoding.The method first divides the urban road network area into grids of equal size and disjoint and extracts the traffic features of the grids from the historical trajectory data.Then it uses convolutional neural networks and gated recurrent units to encode the spatial features and temporal features.MGTTE explores traffic feature extraction and encoding based on road networks.The method firstly generates a graph according to the urban road network structure,and proposes a Multi-Head Graph Attention Network(MH-GAT)based on the graph attention network for the simultaneous extraction of temporal and spatial features.MGTTE also contains a graph node influence factor algorithm designed to select the graph node with a high impact on the trajectory.Both methods use meta-learning algorithms to learn general initialization parameters for deep neural networks and apply the learned parameters to the target city to improve prediction accuracy and model generalization.To show the effectiveness of the method proposed in this thesis,the relevant experiments are conducted on real data sets.The experimental results show that the travel time estimation method proposed in this thesis can effectively extract traffic conditions and their spatiotemporal correlation from trajectory data.Compared with the existing baseline methods,the proposed method also has higher prediction accuracy.Finally,this thesis uses hybrid language programming techniques to design a travel time prediction system,which can predict the travel time of a trip based on the input departure and destination.In addition,the system also implements functions such as map display,city switching,trajectory data loading and display,and route planning.