User Mobility Analysis And Prediction Based On Spatial-Temporal Trajectory

With the rapid development of smart phones and communication technologies,the number of terminals has exploded,and the mobile Internet has penetrated into every aspect of people's lives.Wireless users want high-quality communication anywhere and anytime,which poses a major challenge for 5G and future communication systems.User mobility research can provide basic support for mobile communication in terms of handover management,resource reservation,etc.,and help communication system to deal with the increasing demand caused by the growing traffic and the amount of devices.It can promote service robustness and reduce energy consumption and com-munication delay in communication system.This thesis completes three main tasks in terms of user mobility theory and predic-tion models:First,an empirical analysis of user mobile predictability is performed.The empiri-cal analysis of the mobile predictability theory based on two real-world data sets verifies the validity of the predictability theory.We analyzes the user predictability and theoreti-cal upper bound in the call detail record data set and find that the mean of the real entropy is reduced by 34%compared with the mean of the temporal-uncorrelated entropy and is reduced by 66%compared with the random entropy.The data set dependence of the actual measurement of predictability is revealed.The mobile predictability of col-lege students is analyzed based on the college consumption data set,which shows that there are predictability differences based on the personal information and predictability differences among different data sets.In the existing literatures,predictability study is mostly based on communication datasets,network check-in datasets,etc.However,predictability theory is applied to campus consumption datasets for the first time in this paper,which expands the application scope of predictability theory.Secondly,a single-user mobility prediction model based on diffusion kernel is pro-posed.It is demonstrated that the correspondence between theoretical predictability and actual prediction performance can be modeled by a compound Gaussian function.We proposes a diffusion kernel-based mobility prediction model using spatio-temporal tra-jectory to train.It maps the user's spatio-temporal trajectory to the diffusion process in the specified domain.The prediction results show that the model can achieve about 4%prediction accuracy improvement.For the phenomenon that the achieved prediction performance of users with the same entropy value is different,we testify that the corre-sponding relationship between the achieved prediction accuracy and the predictability entropy value can be modeled by the composite Gaussian distribution.Finally,a similar user-assisted mobility prediction model is proposed,which clar-ifies the significance of similar users for improving prediction performance.The as-sociation analysis algorithm is adopted to mine the user's key mobility patterns,and then we introduce the definition of user movement similarity and calculate the mobility similarity between users in order to find the most similar users.After that,a similar user-assisted mobility prediction model is proposed.The mobility prediction algorithm is designed and the influence on the prediction performance of the two important param-eters of the model,weight and threshold,is discussed.The prediction results testify that the prediction accuracy of the model is 4%higher than that of the single-user model.This paper empirically analyzes the predictability theory,and proves its validity,data set dependence and scalability.The proposed diffusion kernel-based single-user prediction model and similar user-assisted mobility prediction model perform better than the existing models.This paper is significant for mobility theory development and application of mobility.