Research On Moving Target Trajectory Prediction Method Based On Position Similarity And Markov Model

With the rapid development of Internet infrastructure and the improvement of positioning technology,the acquisition of mobile trajectory data has become a reality,and the application of location-based service(LBS)has attracted more and more attention.In today's big data era,the content of trajectory data is complex and diverse.The analysis of potential laws in trajectory data can not only get the movement patterns and laws of moving targets,but also mine the behavior characteristics and interests of targets.Due to the discreteness and large scale of mobile trajectory data,it is difficult to model directly.The Markov model,which is commonly used in trajectory prediction,has low prediction accuracy due to the insufficient use of historical information.Therefore,this thesis focuses on how to divide the trajectory space according to the trajectory data and establish the prediction model of the trajectory end point,the main research contents are as follows:Firstly,in order to realize the serialization of trajectory chain,a dynamic mesh generation algorithm based on space-time density is proposed.In this method,a grid based clustering method is used to divide the sample space of the moving trajectory data.Then,the spatio-temporal characteristics of the trajectory are introduced to divide the cells which are larger than the given threshold hierarchically,and the initial clustering is formed on the divided grid,which avoids the problem of information loss of the trajectory data.At the same time,the spatio-temporal dimension is considered,It better reflects the behavior pattern of the moving target.Data analysis and experimental results show that the method is effective,feasible and time-consuming,which ensures the effectiveness of the trajectory sequence.Secondly,in order to solve the problem of insufficient use of trajectory information in target location prediction by Markov model,and sparse data and over expansion of state space in multi-order model,a moving target prediction method based on location similarity and Markov model is proposed.Based on the traditional Markov prediction model,this method introduces the position similarity factor,calculates the position similar trajectories in the current trajectories and the historical trajectories,finds out the historical trajectories that meet the similarity threshold,and combines the historical trajectories that meet the conditions with the Markov model to predict the future region of the moving target through the probability matrix,The historical trajectory information is fully considered.The experimental results show that the prediction accuracy of the proposed location similarity Markov model can reach 70%,which is 16.8% higher than that of the Markov model.