Research On Group Mobility Prediction Method Based On Mobile Network Data

Mobile behavior is a direct interaction between people and the real world,and large-scale group mobile behavior is the direct cause of many social phenomena and public security problems in the city.Therefore,the research on group mobility is conducive to urban infrastructure construction and public security management.In recent years,with the development of mobile Internet,a large number of user online log data record the multi-dimensional information such as time,space and content of user behavior in detail,and become an important data source for human behavior research.How to effectively use this kind of data,high-dimensional,fine-grained analysis and prediction of group mobile behavior is an urgent problem.Based on the background of mobile network big data,this thesis comprehensively analyzes the regional characteristics of user behavior from the perspective of urban supervision,and forecasts the group mobility of users in urban space.The main work of this thesis is summarized as follows:1.Based on the usage detail records(UDR)of user mobile network,this thesis uses statistical analysis method to analyze user behavior from the aspects of information entropy and spatial correlation.The diversity and heterogeneity of user network access behavior,as well as the regionality and predictability of user mobile behavior are found.Finally,the spatial resolution of UDR data is discussed.2.In this thesis,we propose a regional user behavior feature discovery method based on heterogeneous information network(HIN)representation learning.Firstly,the heterogeneous information network is constructed to comprehensively represent the user's network access behavior and spatial mobile behavior,and the learning method HIN2 Vec is used to obtain the low-dimensional embedding vector of spatial nodes.Then based on the clustering algorithm,the typical regional user behavior patterns are found,and the differences of user group behavior in different functional areas of the city are analyzed.The results show that the embedding vector of spatial nodes can reflect the periodicity of group mobile behavior and the peak of human traffic in each region,which can be used in group mobility prediction.3.A group mobility prediction model EA-STGCN based on spatiotemporal convolution network is proposed.The model mainly consists of two parts,in which the space node embedding vector part takes the space node embedding vector of the HIN of user behavior as the input,and obtains the output value of each node through the fully connected neural network;the graph neural network component takes the convolution network model of spatiotemporal graph as the basic model,and introduces the attention mechanism to dynamically adjust the node connection weight of the graph model In order to capture the dynamic characteristics of the flow direction and tide of the actual mobile behavior of the user group.Finally,the results of the two parts are fused to get the prediction results of the EA-STGCN model.The experimental results show that the EA-STGCN model is superior to other methods in prediction error,stability and other performance indexes.The above research method analyzes the high-dimensional characteristics of mobile Internet users' behavior,finds the typical regional characteristics of user behavior patterns,and uses spatiotemporal convolution network to integrate the time and space dimensional characteristics of group mobile behavior to predict group mobility.The experimental results show that the prediction method based on highdimensional behavior information is better than the traditional method.