A Research On Trip Purpose Recognition And Trajectory Prediction Based On LBS Data

Carrying out residents’ travel surveys and studying the characteristics of residents’ travel have always been important means of urban planning and solving urban problems.With the rapid development of mobile Internet technology and smart phones,the form of residents’ travel surveys is also undergoing changes.From the traditional questionnaire survey to a passive survey that relies on GPS devices and smart phones.Therefore,research on mobile phone positioning data of residents’ travel is of great significance for analyzing residents’ travel behavior and solving urban problems.Based on the LBS data of Weihai residents,the problems of stay point recognition,trip purpose recognition and trajectory prediction were studied separately.First,the principle LBS data,as well as the LBS dataset used,were introduced.The original LBS data had many useless features,chaotic time stamps,spatial distribution is not limited to Weihai,and there were different degrees of trajectory drift.These errors were dealt with separately,laying the foundation for the follow-up work.Then the recognition of staying points based on the spatiotemporal trajectory was completed.Using ST-DBSCAN to introduce the time dimension on the basis of the traditional DBSCAN clustering algorithm,it solved the problem of clustering of trajectory points that are adjacent in space but not adjacent in time.At the same time,by adding clustering rules,possible noise points were eliminated Etc.By identifying the stop point,the start time,end time and specific position of each travel behavior in the travel trajectory were determined.Next,in order to recognize the purpose of the trip from the trajectory data after the stay point recognition,the Xgboost algorithm was used to establish a recognition model.Feature extraction is performed on the processed trajectory data,and the extracted temporal features are combined with the spatial characteristics of land use,POI and other spatial features and the social attributes of residents to construct a feature vector sequence.In the experimental stage,a general trip purpose recognition model was trained using travel data from Jingzhou City.This model’s trip purpose recognition results for Weihai residents were basically consistent with the contents of Weihai residents’ travel reports,proving the accuracy and versatility of the model.Finally,the bidirectional LSTM model was used to complete the trajectory prediction problem.In terms of features,the features of trip purpose obtained by trip purpose recognition were innovatively added.The experimental stage compared the prediction effects of the model used in this thesis with other features and other models.The experimental results proved the effectiveness and accuracy of the method used in this thesis.