Mode Choice Analysis With Incomplete Population Attributes Based On Bayesian Network

Discrete choice model has been widely used in the study of travel mode choice modeling.Travel mode choice is also an important part of transportation planning and decision-making,which can effectively adjust the urban traffic structure and relieve the urban traffic pressure.Due to the uncertainty of resident trip survey,the population attribute data of the respondents can not be obtained completely.It is difficult to deal with this kind of problem by using maximum likelihood estimation.In the meanwhile,with the rapid development of Internet and intelligent transportation technology,more and more traffic big data are applied to the analysis of travel behavior.However,these data lack the attribute data in demographic and socioeconomic level.This paper reviews the development of discrete choice modeling and its application in research and practice at home and abroad,then summarizes the existing methods to deal with missing data.In addition,the basic concepts of Bayesian network model is introduced,and the development and application in the field of transportation is summarized.Bayesian network model is a powerful tool to learn the structure of population attributes,which can establish the correlation of population attribute variables in the case of small or incomplete samples.On the basis of theoretical analysis,this paper puts forward the method of travel mode choice with the condition of missing data,establishes the mode choice model based on Bayesian network,studies the structure learning method combining prior knowledge and constraints,and applies EM algorithm to realize the parameter learning of Bayesian network under the condition of missing data.In order to enhance the interpretability of the model and study the factors affecting the choice of travel mode,a travel mode choice algorithm integrating Bayesian network model and random utility theory is constructed.Travel mode choice node is set as Soft Max node.The probability distribution of missing data variables is obtained by Bayesian network inference,and applied to the construction of discrete choice model.Based on the 1998 swiss metro travel survey data,the structure and parameters of the model are expressed,and the prediction performance of the model is evaluated.Through the resident trip survey data in Chongqing in 2018,this paper constructs and evaluates the model algorithm proposed in this study,and analyzes the temporal and spatial distribution characteristics of Chongqing residents’ trips.A Bayesian network-based travel choice modeling method considering missing data is established.The results show that the model has good log-likelihood value and can consider the correlation between individual attributes.In the case of different missing variables and different degrees of missing data,the proposed models can still get good estimation results and prediction accuracy.