Research On Next Location Prediction Algorithm Based On Similar Behavior Of Mobile Objects

In recent years,with the continuous miniaturization and wide application of positioning technologies and supporting GPS devices,the acquisition of trajectory data has become more and more convenient,and the data mining technology based on historical movement trajectories has been promoted.The existing position prediction studies are mostly based on frequent pattern mining or neural network methods.These prediction methods have limited considerations on the temporal and spatial properties of data and the temporal regularity of active trajectories,and the lack of multi-dimensionality of trajectory data in similarity measurement methods.To solve the above problems,this paper proposes a research on the next location prediction algorithm based on the similarity of mobile object behavior.Firstly,based on the multi-dimensional features of trajectory data,a similarity metric algorithm based on three-dimensional distance function is proposed.From the perspectives of horizontal,vertical and angle,the distance between the trajectory,the gap between the length and the difference between the directions are calculated.Compared with other traditional metrics,the clustering effect of the method is more compact and reasonable.Secondly,a time grid clustering method based on time axis is proposed for each sampling position.The time of day is divided into equal-sized time segments,each segment is a time grid,and then the time stamps of the trajectories are allocated to the corresponding time grids,and the time grids with similar movement patterns are clustered.Furthermore,the similarity between each type of trajectory in the time grid clustering result is calculated by using the similarity measure formula of three-dimensional distance function,and clusters with the same roles,interests and laws are obtained,thereby using the moving data of the similar mobile object groups to improve the accuracy of location prediction.Finally,combined with the moving mode transfer characteristics and trajectory characteristics of moving objects,a hidden Markov prediction model based on the behavior similarity of moving objects is proposed.Firstly,according to the time stamp of the location of the mobile objects,the time grid category to which it belongs is determined,and then the trajectory similarity with other mobile objects is calculated to determine the trajectory class to which it belongs,so as to train the HMM prediction model for prediction.The experimental data came from the GeoLife data set of Microsoft Research.The experiment verifies that hierarchical clustering based on three-dimensional distance has the smallest DB index value and the largest DVI index value,which can achieve more compact and reasonable clustering effect.The coverage of HMM prediction model based on similar behavior of mobile objects tends to 1,and the experimental accuracy is better than other prediction models to predict the next location.