Research On Travel Characteristics Identification Of Expressway ETC Passenger Car Users Based On Combination Model

Along with the proposal of the goal of building a country with strong transportation network,the scale of China’s transportation infrastructure construction has expanded rapidly,forming an expressway network covering 95% of the country’s population,and the coverage rate of expressway ETC users has exceeded 80%.At the same time,as the holdings of passenger cars is rising year by year,passenger cars have become the main part of the expressway vehicle type composition,and their travel purposes are increasingly rich,and the heterogeneity of travel behavior is more and more prominent.Targeted analysis of the travel characteristics of expressway passenger cars,to grasp the rules and needs of their travel,accurate classification of feature groups and individual vehicle identification is the basic work of improving expressway traffic operation status and upgrading travel service.This thesis analyzes the spatiotemporal characteristics of expressway ETC passenger car users through deep mining of expressway ETC data,and finds that there is a group effect in the travel characteristics of ETC passenger car users.In order to thoroughly study the group similarity and individual attribution,this thesis combines the user classification model based on improved K-means clustering algorithm with the feature group recognition model based on BP neural network algorithm to complete the travel feature identification study of "group classification-individual identification" for expressway ETC passenger car user based on selected travel feature indicators and comprehensive analysis of the defects of the basic algorithm.The main research contents are as follows:(1)Spatio-temporal characteristics analysis of passenger car travel.To deeply analyze the vehicle travel spatio-temporal parameters based on the pre-processed ETC data,it is found that the distribution of its travel spatio-temporal characteristics is not balanced,and there are 20% of high frequency travel vehicles providing nearly 60% of the expressway traffic data;other than that,there are differences in the performance of the characteristics of working days and non-working days,in which the travel time,travel distance,and travel length are the most obvious ones;in the spatial distribution of traffic flow,the differences between the toll stations of each expressway are quite large,but the traffic occurrence and attraction of the same station are basically equal.(2)Research on the selection of characteristic indicators of passenger car travel.Based on the results of the spatial and temporal characteristics of vehicle travel,the characteristic index system is initially determined and the process and method of extracting each index are developed,and after refining,four travel characteristic indexes including the number of travel days per month,the average distance of a single trip,the travel preference during peak hours,and the travel preference on weekends and holidays through the index rationality judgment method is finally discovered.(3)The construction of an expressway ETC passenger car user travel feature recognition model based on the combination of cluster partitioning model and deep learning recognition model.In the clustering model,the initial distance threshold of Canopy algorithm is determined by defining the distance distribution histogram to obtain a more reasonable initial clustering center,and then the K-means clustering algorithm is used to iteratively solve the initial clustering results.Finally,the ant colony algorithm is used to further improve the global optimality of the results and obtain more accurate user feature group clustering results.In the part of deep learning recognition model,the BP neural network algorithm with high learning efficiency is used to build an expressway ETC passenger car travel feature group recognition model to achieve the accurate recognition of groups which individual vehicles belong to.(4)The ETC data of Chongqing passenger cars in two months are selected for empirical research.It is verified that the efficiency of model iteration is improved by more than 26.4% after pre-clustering using Canopy algorithm;the accuracy of clustering results is improved by 23.17% after the optimization of ant colony algorithm compared with the basic K-means algorithm.The user classification model based on improved K-means clustering algorithm is used for user classification,which are six groups,to be defined as travel groups,long-distance travel groups,private affairs travel groups,public business travel groups,commuting travel groups,and sporadic travel groups.The feature group recognition model based on BP neural network algorithm is used for group recognition.The results show that the accuracy of model recognition is 95.23%.