| With the development of Internet technology and the popularity of mobile terminal devices,the volume of communication data is explosively increasing.These data have great potential value,providing reliable data guarantee for studying urban space theory and mining human mobility patterns.The rational division of urban areas and understanding of the links between areas are of important value to urban planning.Clarifying user travel demand can alleviate the burden of the transportation system and improve the traffic environment,and provide an important basis for fonmulating the scheduling strategy of the traffic control system.Based on the data of user mobile network,the thesis studies the regional division and regional links,and user travel requirements,as follows:(1)Regional division and analysis of regional connectionThe method of feature extraction and clustering algorithm are studied to divide the urban area,and a related model for analyzing the connection between regions is proposed.Firstly,the time series of base station traffic is constructed,and the 54-dimensional features of the time series are extracted from the statistical,time domain and frequency domain.Secondly,the k-means++ algorithm is used to obtain the functional categories of the base station.Combined with the geographic location characteristics of base stations and the functional categories of base stations,the k-means++algorithm is again used to divide the urban area into 400 functional areas.Thirdly,the Baidu map API is called to obtain the poiint of interest(POI)information of each area,and the function of each area is determined.Compared with the results of the thesis,the average accuracy of the partition is 77.7%.Finally,considering the migration amount of area and the distance between areas,a Newtonian gravity model based on population migration is proposed.According to the model,the connection between regions is analyzed,and on this basis,the importance of the region is studied.(2)Prediction of user travel demandThe models of statistics and machine learning are studied to predict travel demand of users.Firstly,the user travel time series is constructed for each route.In order to deal with the complicated road structure in the city,the thesis proposes to divide the route into three categories according to the frequency of travel of the crowd:fr-equent routes,ordinary routes and sparse routes.Secondly,the Autoregressive Integrated Moving Average Model(ARIMA)in statistics is constructed for all routes.The sliding window method is used to process the user travel time series,and the training set and test set are constructed.Three machine learning models are established:Support Vector Regression(SVR),Gradient Boosting Decision Tree(GBDT),random Forest(Random Forest,RF).After the four models are established,they are used to forecast the travel demand of users,that is,to predict the user travel volume of each route in each period.Thir-dly,mean absolute error(MAE)and root mean square error(RMSE)are selected as eror indicators to compare and analyze the prediction results.The results show that the prediction performance of the GBDT model is optimal.For GBDT,the average MAE and average RMSE of the frequent route,common route and sparse route are(1.479,2.132),(1.549,2.156)and(1.278,1.745).Finally,the visual display of the prediction results is realized. |
