Location Prediction Based On Mobile Phone Signaling Data

With the development of mobile communication technology and the popularization of the Internet,people are becoming more and more dependent on mobile phones,resulting in a large amount of mobile phone signaling data,and the user location information contained therein has important value.In recent years,location-based Services are booming,such as traffic management,city planning,shop recommendation,infectious disease prevention and control,etc.Effective location prediction is the key to achieving the above application scenarios.In the past,most studies on user trajectory location used GPS positioning data,because the data is difficult Therefore,the limitations of the research are relatively large.The mobile phone signaling data has the advantages of real-time,completeness,and full coverage of travel time and space.Therefore,the use of mobile phone signaling data to predict the user's location can provide convenience for various services.Mobile phone signaling data has problems such as uneven sampling,drift and abnormal handover due to signal instability;in addition,different positioning technologies adopted by operators lead to different positioning accuracy,so it is difficult to obtain fine-grained movement trajectories.In response to the above problems,This article explores the semantic trajectory of the user,that is,the trajectory sequence composed of the staying point of the user's activity,and predicts the next location that the user may go to.The main work of this article is introduced as follows:Firstly,design the corresponding cleaning method for various noise location points in the mobile phone signaling data.After removing the error location,design a cyclic and iterable data cleaning method for the redundant location information that is still not conducive to the extraction of the stay point.Then respectively use the extension The time dimension ST-DBSCAN clustering algorithm and the time threshold method with the base station cell as the stay point extraction unit extract the stay points,and the stay points are connected into a user trajectory position sequence.In order to obtain the semantic features of the user trajectory,the interest in the base station cell is used Point information to each base station cell is annotated with functional attributes,such as catering services,scenic spots,science,education and culture,to generate a trajectory sequence containing semantic features.Finally,this paper establishes an LSTM position prediction model that strengthens the semantic correlation between locations,in the LSTM network layer Previously,an embedding layer was constructed to convert sparse one-hot position codes into position embedding vectors to strengthen the semantic association between positions.Experiments have shown that the data cleaning method in this paper can effectively identify noisy data and has a good data settlement effect;ST-The DBSCAN density clustering algorithm is not suitable for the cell phone signaling data of the CELL-ID positioning method in this article,and the simple time threshold method can extract the stay points;compared with frequent pattern mining and Markov model,the LSTM position prediction model in this article is accurate The rate is high and it can achieve a good prediction effect.This article verifies the feasibility of using mobile phone signaling data to predict the user's location,which has a certain reference value for traffic management,advertising,and infectious disease prevention and control.In order to obtain the semantic features of the user's trajectory,this paper uses the Tyson polygon method to divide the coverage of the base station cell,annotates the base station area according to the captured POI information in the base station cell,and then adds semantic features to the track position sequence.Finally,this paper establishes an LSTM location prediction model that strengthens the semantic association between locations to predict user locations.An embedding layer is constructed before the LSTM network layer to convert sparse one-hot position codes into position embedding vectors to strengthen the semantic association between positions.Experiments show that this method can effectively predict the location that users may reach in the future,and verifies the feasibility of using mobile phone signaling data to predict the location of users.The work done has certain reference value for traffic management,advertising,and infectious disease prevention and control.