Research On Travel Pattern Recognition Method Based On Mobile Data

With the continuous development of urbanization,the field of intelligent transportation has become more and more hot,and the development of new technologies has brought a series of new topics that people need to study.Travel pattern recognition is a hot topic in the field of intelligent transportation at this stage.Travel pattern recognition is essential for transportation researchers,engineers,and government personnel to study the behavior of urban populations and to plan,design,and manage transportation systems.In addition,travel pattern recognition also plays an important role in activity-based modeling.This paper mainly studies the identification of urban population travel patterns.By collecting urban population travel trajectory data,analyzing and cleaning the data,extracting relevant features,and using related algorithms of machine learning and deep learning to identify which type of travel mode a single travel trajectory belongs to.The data used in this article is mobile data,including GPS trajectory data and mobile phone signaling data.In addition to the above two,it also uses related geographic information data,including base station location data and subway station latitude and longitude data.The first is mobile data analysis and processing.This article uses the open source Geolife project data from Microsoft Research Asia to conduct research.The form of the project data is GPS trajectory data.This article first matches the label file in the data with the GPS trajectory data file so that each piece of trajectory data has a corresponding travel mode label.Some unnecessary patterns in the 12 types of travel modes of the original data are eliminated,and finally data samples of the six types of travel modes are retained,and a total of 3412 samples are obtained.Then,22 related features were extracted from all data samples.In addition to GPS trajectory data,this paper also uses mobile phone signaling data for related experimental research.For mobile phone signaling data,its coarse granularity and low location accuracy result in this type of data is not suitable for direct training of travel pattern recognition models.Therefore,the mobile phone signaling data experimental part adopts transfer learning ideas.GPS trajectory data to train the model,and then migrate the model to mobile phone signaling data for model verification.Among them,because the original mobile phone signaling data lacks the support of data tags,it cannot be directly used for related experiments.In this regard,this article uses the trajectory data of taxis in Changchun,bus trajectory data in Changchun,and base station location data in Changchun to convert the trajectory data of two types of travel methods into mobile phone signaling data based on certain rules.Adopt the transfer learning idea,use GPS trajectory data to train the model and verify the model,and then transfer the trained model to the mobile phone signaling data for model verification based on the mobile phone signaling data.Among the six sample travel modes after processing,the subway travel mode has different characteristics from the travel modes of other species.The starting point and the ending point of the subway travel mode have the characteristics of being closer to and near the subway station.Since most subway lines operate underground,the signal loss is severe.Coupled with the fact that subway trains are enclosed metal bodies,GPS signals may also be incomplete.This results in no GPS points or only a few GPS point records during a subway trip.These special features make the recognition of subway travel modes different from those of other modes of transportation.Therefore,for the subway travel mode,this article uses the rules based on the distance between the start point of a trip section and a subway station to be less than the critical distance Travel modes are identified.The recognition accuracy of the final result is 0.828897.For the overall classification of the six types of samples in the travel mode,the experimental data samples have the characteristics of high data quality,considerable sample numbers,and obvious sample discrimination.Therefore,this paper also uses a deep factoring machine algorithm based on deep learning to use the previously extracted relevant features to classify the six types of travel patterns of GPS trajectory data.For the transformed mobile phone signaling data,the same characteristics as the GPS trajectory data are extracted,and then the trained model is used to perform model verification on the mobile phone signaling data samples.Finally,for GPS data,the overall accuracy rate reached 0.6735,which is better than the multi-layer perceptron model.Finally,this paper uses a machine learning-based random forest model,LightGBM model,and uses the same processing procedure as the deep factorization machine model to conduct related research on travel data recognition methods based on mobile data.For GPS data,the accuracy of the LightGBM model reached 0.90776,and the accuracy of the random forest model reached 0.855051.The experimental results are significantly better than those of the support vector machine model.For mobile phone signaling data,relevant experiments and experimental results are also compared.