Prediction Of Passenger Travel Mode Based On Transfer Learning And Deep Structured Semantic Models

Nowadays,urban transport departments optimize the structure of public transport system by adding and adjusting public transport lines to cope with the increasing traffic pressure.In order to effectively evaluate the effect of network optimization,it is necessary to accurately predict the passenger travel mode.The traditional method of travel mode prediction is to use logit model and its improved model.There are two main problems in this method: it is difficult to obtain personal information data,and it is difficult to describe the nonlinear relationship between variables.Therefore,this paper extracts the historical travel records of passengers from the massive IC card data,and extracts the characteristics of various travel modes under different travel OD from the structured traffic data such as GPS data and static line information table.By exploring the correlation between historical travel records and travel mode characteristics,a passenger travel mode prediction model is constructed.The specific research contents are as follows:(1)A passenger travel mode prediction model based on transfer learning is proposed.When the bus and subway are on the same line,the different number of stations on the same line has an impact on the choice of passenger travel mode.In this paper,according to the number of stations on the same line,the travel OD is classified.And the method parameter transfer is adopted using the parameter of auxiliary set to assist training target set and to improve the accuracy of passenger travel mode prediction.The validity of the model is verified on the Xiamen public transit passenger IC card dataset.(2)A passenger travel mode prediction model based on deep structured semantic model is proposed.There are differences in the choice of travel mode between different passengers,so it is necessary to learn the preference of different passengers through a large number of historical travel records.Therefore,this paper proposes a passenger travel mode prediction model based on deep structured semantic model,which takes individual passengers as the basic unit to analyze,uses deep learning to learn the relationship between features,and uses deep structured semantic model to learn the interaction between passenger features and travel mode features.The input end of the model is divided into passenger input features and travel mode input features.The two parts of input are extracted respectively,and then the two parts of features are combined to extract the interaction between passenger features and travel mode features.The validity of the model is verified on the Xiamen public transit passenger IC card dataset and London travel demand survey dataset.