Research On Trajectory Prediction Method Based On Machine Learning

With the rapid development of the Internet and positioning technology,intelligent mobile devices have gradually become an indispensable part of life,generating massive trajectory data.Domestic and foreign scholars have used these data to conduct different research.As an important part of trajectory research,trajectory prediction has a wide range of application scenarios,which can provide users better services and provide important basis for decision-making of government and enterprises,such as various location-based user services,crowd congestion early warning,network resource allocation and mobility management.These will help the harmonious development of society and achieve the growth of economic benefits.In recent years,research on traj ectory prediction has become popular,and related methods have been proposed and applied.However,many current studies have mined the track points represented only by a series of incomprehensible numerical labels,and the extraction of the stay points are also lack of time consistency.On the other hand,conventional trajectory prediction models do not make full use of the trajectory context information,and only predict the position of the next stay points,so the prediction accuracy of research results can be improved.Based on the above problems,this paper studies the trajectory prediction method based on machine learning.The completed works are as follows:(1)The processing method of trajectory information is studied.The trajectory data is transformed into more understandable semantic information through a series of processes,such as traj ectory preprocessing,spatial information extraction,time and direction information addition.A spatial information extraction method based on multistage clustering is proposed.Firstly,based on the spatio-temporal coherency extended stay points extraction method,The sliding window and the region coherency expansion algorithm are used to extract the stay points with relative temporal and spatial consistency.Then,based on the heuristic growth clustering method,the stay area points are extracted.After that,the stay area points are merged with the move points.Finally,they are converted into the place name in life,so that the trajectory data can represent the information of the user's activity place and moving process.(2)The traj ectory prediction method based on deep learning is studied to extract the feature vectors of the processed trajectory information and build a prediction model,so as to make full use of the context information and improve the prediction accuracy.In this paper,the Word2Vec model in natural language processing is used to transform spatial information into word vector,combining time and direction features to construct the feature vector sequence.According to the timing relationship inside the feature sequence,the next position of the trajectory is predicted based on the LSTM model and the bidirectional LSTM model in the deep learning,including the stay area points and the move points.(3)The simulation experiment is carried out with GPS data in real life,and the model performance under different parameters is analzed.Experimental results show that the proposed methods in this paper can effectively predict the next location of the traj ectory.The research results of this paper can better represent the semantic information of trajectory and improve the prediction effect,which is effective and practical.