Research On Vehicle Travel Group Identification Method Of Urban Arterial Road Based On License Plate Recognition Data

With the rapid development of economy in my country,the pace of urbanization has accelerated,and the population of motor vehicle has grown rapidly,which has led to a further intensification of the contradiction between people,vehicles and roads.As the lifeline of urban traffic,urban arterial road is an important part of urban traffic system.It often undertakes a lot of traffic travel,and traffic congestion is more concentrated in urban arterial roads,seriously affecting the normal life of urban residents.Conducting classification research on vehicle travel group of urban arterial road can help identify the traffic operation laws,which can provide a basis for traffic management department to formulate targeted traffic demand management policies,and it is an important measure to ensure the normal operation of the urban transportation system.Based on the license plate recognition data of Qingdao Jiaozhou Bay Tunnel,this thesis conducts data preprocessing and analysis of the overall travel characteristics of vehicles,constructs a multi-dimensional and reasonable travel characteristic indicator system,and implements vehicle travel group identification based on improved Kprototypes algorithm and Particle Swarm Optimization-Gradient Boosting Decision Tree(PSO-GBDT)algorithm.The main research contents of this thesis are as follows:(1)Data preprocessing and analysis of overall travel characteristics of vehicles.Detailed processing steps and processes to address the problem of erroneous-missing recognition of vehicle license plate number and redundancy in the original data are proposed to obtain basic data that can be used for research.The overall travel characteristics of vehicles are analysed from the distribution of traffic volume,vehicle type and location,and daily travel frequency.(2)Construction of vehicle travel characteristic indicator system.The travel characteristic indicators are extracted from dimensions of travel intensity,travel time and travel habits to comprehensively describe vehicle travel behavior.The distribution law of indicators is studied to preliminary identify the law of vehicle travel behavior.The correlation analyses between numerical indicators,between classified indicators and numerical indicators,and between classified indicators are conducted and the redundant indicator has been eliminated based on the result of correlation analysis.A travel characteristic indicator system is constructed,which includes 8 indicators,such as travel days,standard deviation of weekly travels,average daily travels,standard deviation of first travel time,and repetition rate of travel time-space patterns.(3)Vehicle travel group identification based on improved K-prototypes algorithm and PSO-GBDT algorithm.In terms of vehicle travel group division,on the basis of considering the characteristics of travel characteristic indicator system,the K-prototypes algorithm that can effectively handle the clustering problems of mixed type data is selected,and improvements are made to address the shortcomings of the original Kprototypes algorithm,including using the standard deviation coefficient method to determine the indicator weight in the calculation of numerical attribute dissimilarity,and introducing the density peak clustering algorithm to select the initial clustering center.And then the clustering experiments on public data sets are conducted,the results show that compared with the original K-prototypes algorithm,the improved K-prototypes algorithm in this thesis has improved accuracy,precision,and recall by 16.22%,14.59%,and 15.76% respectively.On the basis of significantly outperforming the original Kprototypes algorithm,the improved K-prototypes algorithm in this thesis has achieved better clustering performance than the other two improved K-prototypes algorithms.In terms of vehicle travel group recognition,Particle Swarm Optimization(PSO)is introduced to optimize key parameters of GBDT algorithm,and the PSO-GBDT algorithm for recognition is constructed.And then the recognition experiments on public data sets are conducted,the results show that PSO-GBDT algorithm has achieved high recognition accuracy,which is superior to GBDT,Random Forest,Logical Regression,Adaboosting and Xgboost algorithm.In the case of close recognition accuracy,the time cost of using PSO algorithm to optimize key parameters of GBDT algorithm is significantly lower than that of using cross validation method.(4)Experimental analysis.In terms of vehicle travel group division,based on the travel characteristic indicator data,cluster analysis of vehicles based on improved Kprototypes is conducted.The result shows that there are five typical types of vehicle travel groups on the research road: high-frequency commuter groups,low-frequency commuter groups,operating groups,daily travel frequency stable groups and ordinary groups.In terms of vehicle travel group recognition,based on the travel characteristic indicator data with classification,the PSO-GBDT algorithm is used for vehicle travel group recognition,and the result shows that the PSO-GBDT algorithm performs well in vehicle travel group recognition,and the recognition accuracy reaches 98.03%.In summary,this thesis forms an identification method for vehicle travel group of urban arterial road based on license plate recognition data,which has achieved effective division and accurate recognition of vehicle travel group,and can provide theoretical method support for traffic management departments to formulate relevant demand management policies.