The Deep Learning Model For Trip Distribution Prediction Considering Trip Heterogeneity

The trip distribution matrix reflects the spatial distribution characteristics of traffic demand,is also an important basis for traffic planners in traffic planning,infrastructure planning and traffic impact assessment.In the actual trip distribution data,people’s trips have a strong heterogeneity,and this heterogeneity is intuitively manifested that the trips obey power-law distribution,which is also known as the Two-Eight law phenomenon.Statistical physicists call the phenomenon that obeys a power-law distribution as a scalefree phenomenon,that is,most events have a small probability of occurrence,only a few events have a very high probability,and the distribution curve obeys a long-tailed distribution.It can be found from the data of people’s trips that the distribution of trips also obeys the power-law distribution.The OD(Origin-Destination,the origin and destination of trips)with small values of trip accounts for a large proportion,and the ODs with large values of trip accounts for a small proportion.In this paper,the OD pairs with large amount of trip are defined as large values of trip,and the OD pairs with small amount of trip are defined as small values of trip.The threshold for dividing large and small values of trip is calculated according to the cluster center algorithm.Although the proportion of large values of trip is very small,the large values of trip are the main factor affecting traffic planning.Problems such as traffic congestion in cities,crowding on buses and subways,overloading of scenic spots are closely related to the large values of trip.However,traditional trip distribution prediction models(such as gravity model,radiation model)and existing trip distribution prediction machine learning models only focus on the overall trip distribution prediction accuracy,do not consider the extreme heterogeneity of trip distribution,that is,do not pay attention to the large values of trip prediction accuracy.In response to this problem,this paper establishes a trip distribution prediction deep learning model(The Deep Learning Model for Trip Distribution Prediction Considering Trip Heterogeneity,TDPCTH)that considers trip heterogeneity,that is this model considers more about the accuracy of large values of trip.Not only that,this paper also draws on the new ideas of researchers on trip distribution prediction,that is considering the influence of neighboring traffic zones on trip distribution when establishing a trip distribution prediction model.The TDPCTH model takes Chengdu as a case study,and uses the influencing factors of the trip distribution about traffic zones and the influence factors about the trip distribution among the traffic zones as the input to predict the trips between traffic zones.The traffic zones are obtained by dividing the map of Chengdu into grids.The input data includes the residential population,the working population,the number of various points of interest,the area of various areas of interest,the total building area,the distance from the nearest subway station,the spherical surface between the traffic zones.the minimum trip distance and the minimum trip time.On the basis of above data,combined with trip distribution data extracted from mobile phone signaling data,the model is trained and tested.In the process of testing model,this paper compares the prediction accuracy of TDPCTH model with ordinary machine learning models and traditional models.The results show that the prediction accuracy of TDPCTH model exceeds that of the ordinary machine learning models and the traditional models,and TDPCTH model can effectively improve the prediction accuracy of large values of trip.