Research On Vehicle Trajectory Prediction And Missing Information Compensation Method Based On Bayonet Data

In the data analysis of intelligent transportation,the moving path of the vehicle,that is,the vehicle trajectory.Forecasting the vehicle trajectory has become a hot research topic.It aims to apply intelligent information technology to the transportation system,and research and analyze the user behavior,traffic network characteristics and other information to provide users with a more efficient travel experience.However,the complexity of the traffic network structure,the diversity of vehicle trajectory characteristics,and the lack of data in the historical trajectory data bring difficulties and challenges to the vehicle trajectory prediction task in intelligent traffic.How to accurately analyze the road network structure,explore the vehicle trajectory selection law,and compensate for the missing trajectory data is a difficult problem to be solved.In order to solve the related problems,this thesis studies from the following two aspects: combining the vehicle behavior patterns,the spatial and temporal correlation between path points and other influencing factors,the short-term trajectory prediction method is studied;combined with trajectory similarity and user similarity.The missing data compensation method was studied.The main research work and contributions of this thesis are as follows:1.Aiming at the complex spatial and temporal correlation of traffic network and the diversity of vehicle behavior in intelligent transportation,a short-term vehicle trajectory prediction method is proposed and designed.First,the word embedding idea is introduced to analyze the temporal and spatial correlation of the road network.Second,this thesis uses the unsupervised feature learning ability of the deep belief network to achieve the purpose of extracting the local spatial characteristics of the vehicle trajectory.Finally,aiming at the time series characteristics of the trajectory,this thesis uses the linear combination of the current trajectory set in the feature space of the road network to predict the short-term vehicle trajectory.Aiming at the problem of the diversity of the paths under the same destination in the road network,the weight clustering method is used to optimize the results.2.For the data loss problem existing in the actual trajectory data,first,according to the user's time series trajectory information,each trajectory is divided into time segments to find key nodes,thereby dividing the trajectory into key point sequences.Second,this thesis introduced editing distance measures the trajectory similarity between key sequence sequences,and proposed a user similarity calculation method based on trajectory time characteristics and correlation to measure the similarity between users.Finally,this thesis combined with trajectory similarity and User similarity,missing node completion for each track.Lastly,the experimental verification is carried out through the city's real passing data set.Experiments show that the trajectory prediction method adopted and optimized in this thesis can combine the road network structure and space-time characteristics well,and predict the trajectory of the vehicle in a short time.The method can also make full use of trajectory similarity for the data loss problem existing in the data set.The user's similarity effectively compensates for the missing trajectory.