Urban Road Travel Time Prediction Method Based On Deep Learning Of Taxi Spatio-temproal Trajectory

Road navigation is an important tool for people to travel in their daily life.Choosing a right navigation route depends on the travel time of this route and the reliability of reaching the destination within this period of time.Due to the wide coverage,long running time,and large amount of data,using spatio-temproal trajectories of taxis to predict the travel time of urban roads has become an important research content of intelligent transportation systems at home and abroad.But there is a lack of sufficient accuracy and reliability in current prediction methods for travel time.In response to the issue,deep neural network is used to predict the travel time of taxi trajectory in this paper.And the main works are as follows:(1)This paper researched data preprocessing and map matching methods.Using coordinate transformation and Hidden Markov Models to complete the map matching process between taxi's GPS track points and electronic map road network.(2)This paper studied the operation of taxis and travel of residents in Chengdu based on the trajectory data.And analyzed the number of passengers,passenger length,travel volume,travel distance and other indicators.Then,researched the temporal and spatial changes of traffic conditions in Chengdu through the spatial-temporal cluster of taxi travel indicators.(3)This paper proposed a Spatio-Tempro Trajectory Model(STTM),the main structure of which is bidirectional Long Short-Term Memory(LSTM)and residual network.The reliability and generalization ability of the STTM were confirmed by multiple accuracy evaluation indicators on test data.(4)Geoheash coding and DBSCAN(Density-Based Spatial Clustering of Applications with Noise)spatial-cluster were used to deal with the spatial correlation between spatio-temporal trajectory points.The application of word embedding method added valuable features to STTM,and improved extraction method of taxi sample trajectory data.(5)Compared the influence of different factors in the sample trajectory data on the prediction results of travel time.It was proved that taxi ID and matching road ID are more important than other features.Compared the influence of the model cyclic structure on the predicted results of travel time.It was proved that bidirectional LSTM can improve the accuracy of travel time prediction in comparsion with Gated Recurrent Unit(GRU).Experiment shows that the depth space-time trajectory model proposed in this paper can learn the urban road traffic conditions from the taxi trajectory.The mean absolute percentage error of the forecasted value is 6.126%.The mean absolute error reaches 72.416 seconds.The experimental results are superior to other travel time prediction methods.The research results of this paper achieved the path-based travel time prediction between two arbitrary points in city,which has a higher reference value for people to choose the right travel routes.