Research On The Prediction Method Of Vehicle Lane Changing Behavior Based On Driving Trajectory

Nowadays,the number of cars is increasing day by day,and the traffic environment is becoming more and more complex.Lane-changing behavior is one of the most common behaviors in the process of vehicle driving.If the timing of lanechanging is not properly grasped,it is likely to cause trouble to other driving vehicles,resulting in traffic accidents and ultimately affecting the operation of the entire road traffic.Therefore,predicting the lane-changing behavior of vehicles has become a hot research in the field of intelligent transportation.At present,traditional vehicle lane change prediction models are mostly based on physical models,and most of the research data come from driving simulator simulations rather than vehicles driving on real roads.The uncertainty of lane changing behavior in complex traffic environment and the randomness of lane changing behavior under the influence of vehicles in both lanes.This paper conducts research based on the NGSIM vehicle trajectory data set.After smoothing the original data of the primary lane-changing process,the random forest algorithm model is used to extract the appropriate parameter indicators that characterize the vehicle’s lane-changing behavior.The lane-changing behavior prediction model of the network is used for short-term prediction of vehicle lanechanging behavior.The model can also be used to build a decision-making model for lane changing behavior in intelligent driving related research.The main research contents are:(1)Combined with the relevant research of domestic and foreign scholars,on the basis of analyzing the characteristics of lane-changing behavior,the parameter indicators that can characterize the vehicle’s lane-changing behavior are studied and analyzed,and the long-short-term memory neural network model with memory utility is analyzed in this research.applicability.(2)The research uses the LOESS local weighted regression method to preprocess the trajectory data of the lane change process,calculates the vehicle declination angle based on the trajectory and decomposes the driving speed and acceleration,and selects the lane change and lane keeping samples,which are formulated according to the lateral position of the vehicle.Based on this,the 5s data before the vehicle lane change and during the vehicle lane keeping and following are intercepted,and a series of statistical methods are used to analyze the lateral acceleration,the longitudinal speed,the deflection and the deviation before the vehicle lane change and during the vehicle lane keeping and following.The change rules of parameters such as angle and distance between vehicles in front of them are found to be significantly different between the two types of samples.(3)Using the parameter index importance measurement method provided by the random forest algorithm and its advantages of high efficiency and easy implementation,a random forest model is constructed to screen out the longitudinal velocity change,the lateral acceleration change,the time distance of the preceding vehicle,the declination angle and the The time distance between vehicles in the left and right lanes is used as the parameter index of the lane-changing behavior prediction model.Using the memory function of the long short-term memory neural network for the information of each vehicle,based on the data at each moment,the single-layer and double-layer long-shortterm memory neural network lane-changing behavior prediction models were respectively constructed.Excellent,and the accuracy rate is the highest at the planned start time of lane change,thus confirming that the model is a double-layer structure and confirming the start time of lane change.(4)After the lane-changing behavior prediction experiment,the accuracy,recall and precision of the model reached 94.1%,93.5% and 93.3% respectively,and at the beginning of the lane change,the lateral displacement of the vehicle was concentrated between 2.5cm～16.5cm,the longitudinal velocity variation is concentrated in0.25m/s～1.75 m/s,the variation in the distance between the preceding vehicles is concentrated in 5cm～60cm,the time distance between the preceding vehicles is concentrated in 0.45s～2.7s,the declination angle is concentrated in 82°～88°,The time gap between the left front car is concentrated in 0.45s～3.5s,the left rear car time gap is concentrated in 0.45s～2.25 s,the right front car time gap is concentrated in 0.25s～2.25 s,and the right rear car time gap is concentrated in 0.2s～2.2s.The research proves that by analyzing the vehicle lane changing behavior through theoretical analysis,and using a series of statistical methods to analyze the changing laws of the vehicle running state and the surrounding traffic environment before different driving behaviors occur,the parameter indicators that characterize the vehicle lane changing behavior can be preliminarily selected.The prediction effect of the double-layer long-short-term memory neural network lane-change behavior prediction model constructed based on the parameter indicators selected by the random forest model is better,which proves that the lane-change behavior prediction method studied in this paper is feasible.