Research On Predicting Of Car Sharing Traveling Demand Based On Deep Neural Network

With the acceleration of urbanization,people's demand for travel is increasing.The car sharing matches the passengers online through the Internet technology,which solves the problems of traditional taxis and asymmetric information,which effectively meets the passengers' travel needs.In this paper,the in-depth study on the travel demand of the car sharing,using deep neural network technology,can predict the traffic volume and travel distribution of urban residents accurately.In this paper,the pre-processing operation of the car sharing data is firstly carried out.The DBSCAN clustering algorithm is used to time-space cluster the data,and the map matching is used to eliminate the situation that the GPS data is not in the road.Aiming at the wide distribution of urban GPS trajectories and uneven distribution of heat,the Tyson polygon algorithm is used to divide the prediction area,the time and space characteristics of the car sharing travel are deeply understood through big data analysis technology.The deep neural network model was used to predict the two orders of the car sharing travel order and the network travel destination.The main research contents of this paper are as follows:(1)Firstly,data cleaning and format conversion are carried out for the data of car sharing appointment.DBSCAN clustering algorithm is used to cluster OD data of net car appointment points in time and space to improve the accuracy of prediction.Combined with Xi'an road network information,the clustered GPS data are mapped.The Tyson polygon generation algorithm is used to spatially divide the prediction area of Xi'an.(2)Using the big data analysis technology,the time-slicing operation is used to divide the 24 hours a day into equal-length time segments,and analyze the travel time of the network by analyzing the travel time of the vehicle.After the spatial region is divided,the distribution of the traffic volume in the network of each sub-region is analyzed,and the distribution of the cold hotspots in the network is analyzed.(3)The deep neural network model combined with convolutional neural network(CNN)and long-term and short-term memory network(LSTM)is used to predict the travel demand of the network.Through the data analysis tools such as SPSS,the input model of the car sharing travel order forecasting model and the car sharing travel destination forecasting model are determined,and the forecast of the car sharing travel order volume and the car sharing travel destination is realized.This thesis focuses on the research of car sharing travel demand forecasting.In order to make full use of the spatio-temporal characteristics of data,the spatial characteristics of the data are first extracted through the CNN network.Then the time characteristics of the data are extracted through the LSTM network.The above algorithms and models are evaluated by using Xi'an car sharing booking order data and GPS trajectory data.The experimental results show that the depth neural network model adopted in this paper has higher prediction accuracy(PA)and lower root mean square error(RMSE)than the traditional prediction model.