Research On The Attractiveness Of Urban Areas Based On Private Car Trajectories

The evolution of cities are the result of the long-term interaction between human activities and area structure.The travel behaviors of urban citizens are closely related to the characteristics of spatial and temporal structure.Area Attractiveness(AA)distribution because some specific area in the city attract people,which is closely related to human mobility and the rhythm of urban life.With the rapid development of urbanization and motorization,an increasing number of people choose to buy private cars to travel to destination to fulfil their daily travel needs,which leads to the spatio-temporal evolution of urban areas attractiveness and the formation of the aggregation effect of some hot spots in urban area.For policymakers and researchers,understanding the relationship between people's mobility and urban areas will benefit the widespread applications of intelligent transportation and location-based services,such as traffic control,POI and advertising recommendations,parking planning,and more.By observation,human travels show a regularity of spatio-temporal pattern on weekdays,but on weekends,it is more influenced by various factors,such as weather conditions,etc.,hence the destinations sets present vast difference between different drivers,therefore it is difficult to predict,in addition,the stay time of private cars produce an impact on the results of the aggregation effect,but often ignored in previous studies.This paper focus on the research of the variation of urban areas attractiveness and areas aggregation effect in special time based on thebig scale trajectories of private cars.But in the dynamic of urban environment,urban area attractiveness and aggregation prediction have enormous challenges,mainly include 1)how to modeling the urban area attractiveness and the spatial features distribution of aggregation effect,2)how to effectively predict with the fusion of multi-source data,3)how to predict when people travel with random pattern during the holidays.In order to solve the above problems,this paper adopts the Variable Bayesian Gaussian Mixture Model(VBGMM)for the modeling of the spatial distribution of urban area attractive,and LSTM to fit based attractiveness distribution on time series.Besiedes,drop out with experiential probability in the neural network to improve an ability of network generalization so that enhance the prediction effect,accelerate the training speed of neural network at the same time.In the modeling of the aggregation effect of urban areas on weekends,this paper integrates the temporal representation module,spatial representation module and external embedding module.An improved Kernel Density Estimation algorithm based on Log-Cosh loss function is proposed to model spatial features.Log-normal distribution is applied to model stay time,which is integrated with spatial features.Weather conditions are input into the neural network as external features,and GRU model is used to model changes in time series.In order to cope with the low entropy state of aggregation effect,NALU module was added in this paper to enhance the numerical extrapolation ability of the model.Therefore,the main contributions of this paper are as follows:1)VBGMM and LSTM are used to model the spatial distribution and time series of attraction respectively,and Dropout is used to improve the generalization ability of the network and accelerate the training progress;2)Kernel density estimation algorithm based on log-cosh loss function and log-normal distribution were used to model spatial features and stay time,and the space-time attention fusion module was designed to enhance the prediction ability of the model;3)We consider weather conditions as external features embedded,and GRU model is used to model the changes of time series;4)The NALU module is used to enhance the numerical extrapolation ability of the model;5)This paper conducts experiments on the real world big-scale private car trajectories,and the results show that the proposed scheme is superior to the existing research schemes.