A Trajectory Data Segmentation Method And Application Based On Semantic Features

With the popularity of Global Positioning System,trajectory data has been applied to various fields of our daily life.Trajectory segmentation is an important pre-processing in trajectory data mining,which effectively helps us to understand and use trajectory data.However,the traditional trajectory segmentation methods can only be applied to specific application domains and trajectory data sets.Based on the semantic features of trajectory data,we propose a trajectory data segmentation method that can be applied to a variety of applications in this thesis.And,a trajectory semantic annotation framework and a traffic volume prediction model are constructed by combining this trajectory segmentation method.The main contents include:(1)A trajectory segmentation algorithm based on trajectory features,called Moving Pattern Change Detection algorithm(MPCD),is proposed to complete the segmentation task of trajectory data.MPCD segments the raw trajectory without relying on prior knowledge or segmentation criterions based on thresholds.Moreover,by selecting different trajectory semantic features,MPCD can be applied to different trajectory data sets and applications.In the comparison experiment,MPCD has the same distance-accuracy as the comparison method,but the time-accuracy is 2.5% higher than the comparison method.(2)A trajectory semantic annotation framework for the trajectory semantics enrichment is constructed in this thesis.The framework is different from the traditional trajectory semantic annotation methods,which contains complete trajectory data pre-processing and semantic reasoning mechanisms.Specifically,after the trajectory segmentation,the established general ontology model annotates the general semantics of the trajectory from the perspectives of user behaviors and transportations,which provides complete semantic reasoning mechanisms for the trajectory.The experiment exhibits the result of the hierarchical trajectory semantic by analyzing the study case,and makes an interpretation of the trajectory semantics.(3)A traffic flow forecasting model for urban areas is established.Based on the temporal relationship between traffic flow data in different time intervals,the model accomplishes continuous data prediction for traffic flow data between urban divisions.Meanwhile,in order to reduce the pressure of model parameter learning,a parameter learning algorithm is proposed.Under the premise of ensuring the accuracy of model prediction,the batch learning of parameters is completed.And the experiment results show that the prediction error of this model is reduced by 8.36% and 12.20%,respectively,compared with the comparison experiments.By improving the traditional trajectory segmentation,the proposed trajectory segmentation method based on semantic features avoids the disadvantages of using thresholds.Therefore,the segmentation method can be applied to more trajectory applications.Based on the segmentation method mentioned above,we establishes a trajectory semantic annotation framework and traffic flow data prediction model in this thesis,which can be used to annotate trajectory semantic information and predict traffic flow data between urban divisions.