Research On Vehicle Trajectory Prediction Algorithm Based On Driving Intention In Connected Environment

Intelligent networked vehicles can effectively help drivers and intelligent driving systems cope with upcoming road risks by sending self-vehicle behavior information and receiving traffic semantic information from the surrounding environment to surrounding traffic participants.However,the complex and dynamic spatial interaction behavior of the vehicle-workshop in the connected environment changes the attention distribution and decision-making ability of the driver and the intelligent driving system,and the driving behavior prediction model constructed in the traditional environment will no longer be suitable for complex and changeable connected driving scenarios,and it is urgent to propose a more targeted driving behavior prediction method in the connected environment,which is of great significance for improving the driving safety of intelligent networked vehicles and the operation efficiency of the entire traffic flow.This paper focuses on the interactive lane changing behavior of vehicles in the connected environment,and proposes a lane change intention recognition model and a multimodal trajectory prediction model considering vehicle dynamic interaction.The main work contents are as follows:(1)A sample set of vehicle driving behavior in a connected environment was established.The generation mechanism and data differences of vehicle lane changing behavior in connected environment and traditional environment are analyzed in depth,and the NGSIM dataset is selected as the data source for the driving behavior prediction model research in this paper.The position,velocity and acceleration noise existing in the original data are effectively smoothed,and on this basis,the interaction information of the lane changing vehicle is extracted,and the vehicle driving behavior dataset with the characteristics of networked interaction is obtained.(2)A vehicle lane change intention recognition model based on CNN-Bi GRU-Attention in the connected environment was constructed.By introducing1D-CNN,the spatial association of vehicle historical interaction information is mined and deep high-dimensional time series features are extracted.The intention output module based on the attention mechanism is designed,which enhances the model’s ability to pay attention to the important hidden features of each time step of the GRU unit in the process of intent recognition.The results in the test set show that the overall intention recognition accuracy of CNN-Bi GRU-Attention is 93.47%,which is better than the mainstream intention recognition model,which can accurately identify the vehicle’s lane change intention 2s before the vehicle changes lanes,and has good early recognition performance.(3)A multi-modal trajectory prediction model STA-LSTM considering vehicle dynamic spatial interaction is constructed.The interaction relationship between vehicles is modeled by occupying a spatial grid,and the attention mechanism is introduced from the spatial dimension to adaptively match the weight of the influence of surrounding vehicles on the target vehicle,and the attention module is embedded in the LSTM decoder from the time dimension,so that the model can identify important historical hidden features during each trajectory decoding process.The results on the test set show that the RMSE values of the predicted trajectories of the STA-LSTM prediction model from 1s to 5s are 0.46 m,1.15 m,1.89 m,2.84 m,and 4.05 m,respectively,which can effectively predict the intention and driving trajectory of connected vehicles.