Research On Vehicle Moving Trajectory Prediction Based On Deep Learning

Mobile Internet of things and positioning technology provide convenience for continuous acquisition of large-scale trajectory data.How to accurately predict the future position of mobile objects based on historical trajectory data is a hot topic.Due to the sparsity of trajectory sequence data,the inconsistency of trajectory sampling frequency and the complexity of urban traffic scene,it is difficult and challenging to achieve accurate trajectory prediction.How to consider the influence factors of moving trajectory trend change in many aspects,how to consider the road network structure and traffic condition variability in trajectory prediction to improve the prediction accuracy and reduce the sensitivity of trajectory prediction accuracy to sampling frequency are the research difficulties.In view of the above problems,this paper mainly studies from the following two aspects: extracting the dynamic characteristics of vehicle movement behavior,combining with the track sequence data to construct the data feature space,and on this basis,it studies the moving trajectory prediction method;extracting the regional connectivity and the future road condition of the region,combining with the regional position sequence to construct the track feature expression,and on this basis,it carries on the research on the vehicle movement trajectory prediction The location prediction method of mobile area is studied.The main research work and contributions of this paper are as follows（1）Considering the influence of vehicle movement behavior and other factors on trajectory prediction modeling,this paper proposes a vehicle trajectory prediction algorithm based on feature association.The main contents include: 1.Enhance the expression ability of the original trajectory by combining the characteristics of vehicle movement behavior and traffic environment to obtain the multi-dimensional characteristics of the trajectory;2.Do the correlation analysis and importance analysis of the multi-dimensional characteristics to remove the high redundancy and low contribution features;3.Build the trajectory prediction algorithm model based on the trajectory data expressed by feature vector combined with LSTM network.（2）Considering the influence of trajectory sampling frequency and regional road conditions on trajectory prediction modeling,this paper proposes a vehicle trajectory prediction algorithm based on future road conditions of connected regions.The main contents include: 1.Divide the grid according to the distribution of the track,and transform the moving track sequence into the grid position sequence;2.Obtain the regional connectivity according to the connection relationship of the road segments in the road network,and propose a road condition prediction method by embedding the regional connectivity considering the temporal and spatial changes of road conditions and the connectivity structure of the road network;3.Combined with the road prediction output and regional connectivity,the road characteristics of the connected regions are extracted,the feature vector space of the trajectory data is constructed by fusing the grid position information,and the trajectory prediction algorithm model is constructed by combining the corresponding deep learning network.In this paper,the above two algorithms are tested on real taxi trajectory datasets.The experimental results show that the algorithm proposed in this paper has a good effect on the vehicle trajectory prediction.The research results provide a new idea for trajectory prediction,and have a certain reference value for related research work.