Research On Vehicle Trajectory Clustering And Prediction Algorithm Based On Spatiotemporal Data

With the vigorous development of global positioning technology and mobile Internet of Things,a large amount of vehicle trajectory data has been generated.These data has spatiotemporal characteristics and can not only be used for traffic flow analysis and road network planning,but also for vehicle behavior analysis and driving behavior research.How to mine the common characteristics between trajectory data through clustering and realize the prediction of vehicle position in the future based on historical trajectory data is not only a hot spot of current research,but also a key point to realize intelligent transportation and solve urban traffic congestion and safety problems.However,due to the great uncertainty of vehicle motion in complex spatio-temporal scenes,how to effectively mine the spatio-temporal features of data and realize the clustering and prediction of trajectory data has become an important issue.The current trajectory clustering algorithm ignores the impact of timeliness on the trajectory.The traditional trajectory prediction algorithm usually only focuses on the historical trajectory of the target vehicle,cannot effectively extract the spatiotemporal characteristics,and does not make full use of the current road conditions and the interaction between the surrounding traffic participants and the target vehicle.Therefore,from the perspective of spatio-temporal characteristics,the thesis studies the long-distance trajectory clustering and short-distance trajectory prediction algorithms.1.The thesis proposes a trajectory clustering method based on spatio-temporal composite feature similarity.Starting from two perspectives of time and space,this method combines multi-dimensional features,and uses similarity measurement methods based on time window and spatial trajectory feature points to cluster GPS trajectory data.Trajectory clustering comparative experiments were conducted on multiple classic trajectory similarity measurement algorithms using a public dataset,and the results show that the algorithm proposed in this thesis can fully mine multi-dimensional spatio-temporal features between trajectories and significantly improve the effectiveness of trajectory clustering.2.This thesis proposes a Transformer vehicle trajectory prediction model based on spatial attention and convolution social pooling: SACS-Trans.The model takes Transformer as the framework,uses convolutional social pooling layer to extract interactive features of vehicles around the vehicle,uses spatial attention mechanism to obtain spatial attention,and finally integrates vehicle mobility to achieve vehicle trajectory prediction.Experiments on vehicle trajectory prediction were conducted on a public dataset NGSIM with multiple models,and the results show that the model proposed in this thesis performs better than other models in trajectory prediction tasks,especially in long sequence prediction.The trajectory clustering method and vehicle trajectory prediction model proposed in this study can be applied to intelligent transportation systems,autonomous vehicles and other fields,and have broad application value.At the same time,this study provides a new idea and method for mining spatio-temporal features of trajectory data.