Indoor Destination Prediction Based On Bidirectional Recurrent Neural Network Algorithm Research

In recent years,the popularity of GPS embedding equipment and wireless equipment has made the positioning technology develop rapidly and recorded a large number of users ' historical moving information.This information contains a wealth of content,such as the user's day-to-day behavior,hobbies and so on.How to make full use of these information to complete the prediction of the destination has been a hot research topic,and has made a lot of research results.Especially with the rapid development of location-based service,the location-based service has gradually become an indispensable part of daily life,such as target advertising,destination recommendations.Effective destination prediction can greatly enhance the user experience,with high research value and application scenario.Currently,a major method of predicting destination is predicting destination by using GPS positioning technology,the main idea is matching the historical trajectory information with the user moving path and considering the most likely trajectory destination in the historical trajectory dataset is the user's destination.However,GPS positioning technology is mainly used for outdoor space,can not accurately obtain the location of indoor space.And due to the difference between indoor space structure and outdoor space structure,many algorithms of destination prediction for outdoor space can not apply into indoor space.At the same time,many destination prediction do not consider the problem of data sparse caused by different sampling strategy.This paper presents a new algorithm for indoor destination prediction.This algorithm preprocesses indoor space and the sequence of sampling points.The step of preprocessing the sampling points solves the problem of data sparse which causes by different sampling strategies.At last,complete the destination prediction using the preprocessed sampling points.The main work of this paper includes the following aspects:(1)An overview of destination prediction technology.This paper briefly introduces the development and research status of destination prediction,then summarizes the research work of destination prediction,and analyzes the advantages and disadvantages of current destination prediction,at last introduces the application in our daily life.(2)Preprocessing of indoor space.Firstly,we introduce the differences of indoor and outdoor spatial structure,and then describe the algorithm of preprocessing indoor space,which is able to judge the connectivity of indoor grid space by using the historical trajectory dataset.In addition,the indoor space structure is well preserved while the indoor space is divided.(3)Preprocessing of sampling points sequence.In order to solve the problem of data sparse due to different positioning strategies,we propose a method to preprocess the sequence of sampling points,which calculates the spatial and temporal transition probability of different cells and regions by using the indoor grid structure and historical trajectory dataset.And then we preprocess the sequence of sampling points,the preprocessed sequence of points is similar to the original trajectory.This method avoids data sparse caused by different sampling strategies,which ensures that the destination prediction is not effected by different sampling strategies.(4)Destination prediction based on bidirectional recurrent neural networks.At last,we briefly introduce the characteristics of bidirectional recurrent neural network,and adjust bidirectional recurrent neural network to predict destination.Then the preprocessed sequence of sampling points and historical trajectory dataset are applied to bidirectional recurrent neural network to predict the sampling sequence destination.At last,the experimental results show that the algorithm proposed in this paper can predict destination for indoor space,and the results prove the proposed algorithm's accuracy and robust by comparing with other prediction algorithms.