Research On Short-term Trajectory Prediction Of Vehicles For Complex Feature Spaces

In intelligent transportation,how to extract effective information from vehicle trajectory and then make short-term and accurate prediction is a hot issue in current research.Due to the diversity,sparsity and complexity of road topology,it is difficult and challenging to accurately predict the trajectory.How to extract the relationship between nodes from the complex road topology,how to alleviate the problem of trajectory sparsity and how to consider the factors affecting the trajectory trend from a multidimensional perspective are the main difficulties at present.In order to solve the above problems,this paper mainly studies from the following two aspects: firstly,homomorphism compensation is carried out for sparse and missing trajectories by combining the influence factors such as vehicle preference features and node space structure relationship,that is,the original trajectory is enhanced,and then short-term trajectory prediction method is studied.Combined with the potential periodicity in vehicle timing features,the variable trajectory period is modeled and the short trajectory prediction method is studied.The main research work and contributions of this paper are as follows:1.In view of the diversity of vehicle driving process,vehicle trajectory is extracted,and a short-term trajectory prediction method based on preference feature corpus is proposed.First,the generated antagonism network is introduced to learn the trajectory distribution,and then the generated data is used to make up for the loss of the original trajectory and alleviate its sparsity.Then,based on the analysis of the preference features of the trajectory and the spatial features of the trajectory nodes,the representation learning is used to construct the preferred corpus and the feature space of the node structure.Finally,trajectory trend prediction is carried out for different feature vectors,and the prediction trend of multidimensional multi-feature space is fused and the prediction result is formed through maximum pooling.2.In-depth analysis of the potential periodicity in the trajectory timing feature,the vehicle trajectory is extracted by different window modes,and a short-term trajectory prediction method based on variable timing feature is proposed.First,the trajectory is learned by using the graph convolutional neural network to form the overall representation of trajectory information.Then,according to the difference of cycle correlation,the trajectory is extracted by window to form a variable time series cycle dimension,and then the trajectory is represented respectively.Meanwhile,the trajectory is trained by the long short time memory network,and the weight of each cycle is assigned dynamically.Finally,the predicted trajectory trends of each time period dimension are integrated to improve the accuracy of the results.This thesis is verified by the real data set of passing cars in a provincial capital city in China.Experiments show that the trajectory prediction method designed in this study can not only compensate the data,but also combine the features of trajectory preference and the information of road network structure.Meanwhile,the periodic law of the trajectory can be better analyzed.