Factors Affecting Analysis And Prediction Modeling Of The Car Ownership In Residents' Households

With the rapid development of the economy and the improvement of residents'living standards,more and more families are starting to own cars.Convenience is provided by private cars,however,problems such as traffic congestion and environmental pollution have been arisen.Therefore,it is particularly important to explore the factors influencing car ownership and predict its number correctly in order to provide decision-making for city managers.Firstly,the influencing factors of car ownership are searched not only through the relevant research at home and abroad but also combine the actual situation of China.The fourth and fifth Beijing Person Trip Survey data are processed by using Access and MySQL databases,the data is transformed person-based into family-based and then outputted.The geographic location data of public transportation stations is obtained through Gaode map API,and it is matched with the traffic community layer of Person Trip Survey through ArcGIS.Then the coverage rate of 300-meters public transportation stations of each traffic community could be analyzed.The coverage rate data and the database output data are used as the basic data set of this study.The ordered Logit model is used to analyze the various factors which have influenced on the family car ownership.The results show that car ownership is most sensitive to the number of driver's licenses.Secondly,the random forest and support vector machine combination model and the generalized BP neural network model are proposed to predict the number of car ownership in each family with the ideas of machine learning are taken into consideration as well as the characteristics of basic data sets.As for the random forest and support vector machine combination model,the SVM is used to predict the family car ownership on the premise of the basic data is filtered by random forest.As for the generalized BP neural network model,which combined the model with the characteristics of sparsity in the basic data.The method of Factorization Machine has been chosen and the full connection layer(including the Dropout method)is also selected.Both methods are composed of the hidden layer,so that the layer is redesigned than any other theses.The Softmax function is used as the output layer,the cross entropy is used as the loss function to carry back the error,and the Adam method is used to optimize the back-propagation process.By evaluating the above models with different training sets and test set data,it is found that the generalized BP neural network model has better generalization and accuracy.Finally,the macro-model is established in order to predict the number of car ownership in Beijing.The data expansion model is proposed,which used to extend the number of car ownership predicted by the micro level into the whole city level.In this process,a Compertz model suitable for the thousand-person car ownership rate of Beijing is also calculated.By comparing the data obtained by the time series model(ARIMA)with the data expansion model,it is considered reasonable that the number of car ownership in Beijing is 6.642 million in 2035.