Vehicle Speed Prediction Model Based on Traffic Macroscopic Perspective

Accurately predicting the speed of the Surrounding Vehicle (SV) can improve the safety and efficiency of autonomous vehicles' overtaking, lane changes, acceleration and deceleration. This paper proposes a prediction model for the first time to predict the speed of environmental vehicles from the macroscopic view of traffic flow so as to improve its prediction accuracy. Firstly, the NGSIM data set is used as the research material, and the variable of vehicle speed is selected as the output of the prediction model. In the meantime, the inputs to the prediction model are: speed (one time step ahead of the predicted speed), distance from the vehicle in front, and traffic flow speed/density (same time as the vehicle's predicted speed). Secondly, the Spearman correlation coefficient of traffic flow speed/density and vehicle speed is studied with the change of prediction time. The study finds that traffic flow speed/density and vehicle speed maintain a strong correlation. Finally, three models are established by using BP neural network and CNN-LSTM respectively: a) Input speed of vehicle and distance between the vehicle and the front vehicle, without input traffic flow speed/density; b) Input only traffic flow speed/density; c) not only input speed of vehicle and spacing between the vehicle and the front vehicle, but also input traffic velocity and density. By Mean Absolute Error (MAE) and root Mean Square Error (RMSE), it is found that adding predicted macroscopic traffic flow parameters to the prediction model is conducive to reducing the error of predicted vehicle speed, which proves that the accurate prediction of macroscopic traffic flow parameters has a positive effect on microscopic vehicle speed prediction.