Preference Analysis For Shared Autonomous Vehicles Considering Latent Classes

With the development of autonomous driving technology and the popularization of sharing economy,the emergence of Shared Autonomous Vehicle(SAV)has brought tremendous changes to the transportation field.It provides people with convenient,comfortable and economic travel services.Studies show that it has the advantages of alleviating traffic congestion,reducing energy consumption and vehicle emissions,etc.Therefore,SAV has a bright future,but its introduction will inevitably affect people's existing travel behavior.In consequence,it is extremely necessary to find an accurate preference analysis method for SAV to clarify the travelers' mode choice behavior concerning SAV.Based on the literature review,it is found that current preference anlysis for SAV is insufficient in the consideration of factors and individual heterogeneity,and the Latent Class Analysis is gradually being applied to preference analysis.Based on this information,this paper considers more factors,and uses the method combined Latent Class Analysis and Discrete Choice Model to establish the pereference analysis model for SAV considering latent classes,and analyzes the influencing factors and their effects on SAV use,so as to provide scientific basis for the use of SAV in the future.Firstly,the factors and alternatives set are determined,and the RP+SP questionnaire is designed.311 valid questionnaires are obtained by online distribution,and statistical analysis and data processing are carried out.Then the variables are selected to conduct Latent Class Analysis by Mplus software(version 7.4),and data are divided into three classes.Two indexes of valuing category accuracy are 0.983 and 0.967,indicating that the classification is reasonable.Based on the characteristics of each class,they are named the Impulsive and Positive Innovator,the Contradictory and Conservative Innovator and the Rational and Conservative User.The Multinomial Logit model and Mixed Logit model are established for the total samples and various classes respectively.The random parameter distributions of the Mixed Logit model are assumed and tested,and the parameters of all models are calibrated by R language(version 3.4.1).Then,the models named MNL and MNL_IPI,MNL_CCI,MNL_RCU,MIXL and MIXL_RCU are obtained.Combined the results of Latent Class Analysis and above models,the preference analysis models for SAV considering latent class,MNL_LC and MIXL_LC,are gained and the goodness of fit indexes of all models are calculated to valuate models.The results show that MNL_LC and MIXL_LC are over 6.93% and 8.64% better than MNL and MIXL respectively.Finally,parameter results analysis and marginal utility analysis are used to find out the specific impact of each factor on the preference of SAV.The results show that the proposed model considering latent classes shows better goodness of fit and explanation ability than the general model.The analysis results show that the significant factors influencing the SAV use are gender,age,education background,driving experience,vehicle ownership,children,usual travel purpose,understanding depth for SAV,SAV users' type,number of people in the trip,waiting time and average travel cost;Moreover,the significant factors are varied for different classes,and the SAV users' type is the common significant factors for three classes.Based on the marginal utility results,it can be inferred that the increased parking problem for private cars or the longer waiting time of ride-hailing service will significantly increase the probability of using SAV;after the average travel cost of SAV is reduced,the people who use private cars are most likely to change to use SAV;the SAV imitator is the most sensitive to the changes of SAV's attributes,while the innovator is the least sensitive;in addition,the change of SAV's waiting time has the greatest impact on the probability of using SAV than other alternatives specific attributes and the second is the average travel cost.As a result,during the promotion process of SAV,the first consideration is to optimize the SAV's technology to reduce its waiting time or to decrease its price to attract imitators and non-users to use it.