Research On Highway On-Road Vehicle Path Prediction Based On ETC Data

In recent years,the ETC(Electronic Toll Collection)gantries built along highways have not only improved the efficiency of highway traffic,but also recorded massive amounts of data on vehicles traveling on highways.The generation of this massive amount of data has made it possible to construct intelligent highways,with vehicle path prediction being an important part of such construction.Vehicle path prediction can help traffic managers optimize road operations and assist in the operation of autonomous vehicles through vehicle-road coordination.This study aims to use ETC data to predict the path selection of vehicles traveling on highways within10-20 minutes,providing research support for the construction of intelligent highways.This study is based on ETC data from highways.By obtaining the driving trajectory of vehicles through ETC data,this dissertation analyzes and explores the driving trajectories of vehicles,and then constructs a prediction model for on-road vehicle path selection on highways based on ETC data.This dissertation accomplishes the following tasks:(1)ETC data pre-processing and trajectory data extraction were performed.Taking the ETC data generated within one month from the highway in Fujian Province as the research object,we firstly introduce the source and generation of data,analyze the potential problems in the data;secondly,design the data cleaning and repair methods according to the characteristics of the highway ETC gantries;finally select the research vehicles according to the number of vehicle trips and introduce the extraction methods of vehicle trajectories,so as to lay a good foundation for the subsequent chapters.(2)Vehicle trajectory into the service area detection algorithm were designed.Since the vehicles entering the service area will be delayed to reach the next gantry,a vehicle entering the service area identification algorithm is proposed to segment the trajectories that exist to enter the service area section.By combining ETC data and service area data,the driving characteristics of vehicles through the existence of service area sections are analyzed,and the vehicle entry service area detection algorithm is established by XGBoost algorithm.The experimental results show that the prediction accuracy of the algorithm reaches 95%.(3)The ETC gantry feature vector construction method was proposed.In this dissertation,we borrow the word vector representation in natural language processing to transform the gantries in vehicle trajectories into feature vectors.The vehicle trajectories on the highway are used as the training set,and the neural network is used to learn the spatial distribution correlation of the gantries in the vehicle trajectories,and the highway ETC gantries are transformed into feature vectors that conform to the spatial distribution characteristics,and the obtained feature vectors have low dimensionality and perform well in terms of similarity and distance between neighboring gantries vectors.(4)A highway vehicle path prediction model were constructed.Based on the previous work,a neural network model combining GRU and attention mechanism is proposed as a multi-classification problem for vehicle path prediction,in which GRU learns the spatio-temporal correlation in the trajectory sequence and the attention mechanism learns the different influence weights of different parts of the trajectory on the outcome to realize the path prediction of on-road vehicles on the highway.Experimental results on real data show that the model outperforms other models in prediction,with an average accuracy of 96%.