Research On Moving Traiectory Prediction Algorithms

With the popularity of mobile Internet and various mobile terminals,location-based services have become a part of human life style.Various mobile terminal devices have produced a large number of user track data,which contains great commercial value.The research point of this paper is to predict the user's position and behavior based on the user's GPS track.This problem is divided into two subproblems.First,the original trajectory of the user is extracted to obtain the sequence of the user's residence points and the weight of the residence points.Then the user position and user behavior are predicted based on the user residence point sequence and the corresponding residence point weight.At present,in the research of user behavior prediction,part of the emphasis is on mining user information beyond the trajectory,but the data containing this information is less and is not easy to obtain.Actually track itself also contains a lot of information available,the current mining these information can provide more useful help to predict,so a lot of research on trajectory data mining,However,if the user's behavior characteristics are predicted directly,such as speed,residence time,etc.,the regularity of each user's behavior characteristics in time sequence is not strong,and multiple user behavior characteristics are predicted at the same time,resulting in high computational complexity.In addition,the existing prediction tasks,location and behavior are predicted separately,ignoring the correlation between user behavior and user location,resulting in poor prediction effect.The main work of this paper is as follows:The user track information is mined to predict the user's location and behavior simultaneously.In the stage of residence point extraction,heuristic threshold method and logistic regression algorithm are combined to achieve the residence point extraction,and at the same time,the weight of the residence point--the embodiment of user behavior is obtained,which represents the general characteristics of a variety of user behavior information,and solves the problem of poor regularity caused by scattered behavior characteristics.In the prediction stage,the weights of residence points and corresponding residence points are combined to obtain new element sequences,so as to achieve the effect that user position and user behavior can influence each other in the prediction.Then,the new elements are vectorized based on this data,and finally,the user position and user behavior are predicted by LSTM.In order to get the basic processing unit of heuristic threshold method and the sample of logistic regression model conveniently and quickly,This paper divides spatial regions based on geographic grids.Aiming at the problem of residence point splitting caused by space partition,a trajectory merging algorithm is proposed.In addition,a trajectory preprocessing algorithm is proposed to adapt to grid generation,feature extraction and other operations.Finally,through experiments,the feasibility of the algorithm is verified,and the prediction performance is greatly improved.