Research On Traveling Pattern Recognition Based On The Trajectory Information

Travel trajectory pattern recognition in the user query,behavior prediction,interest recommendation based on user's location,privacy protection and municipal traffic planning and several aspects has a wide range of applications.As a result of current identification accuracy can't get the application requirements,so the travel pattern recognition research is the focus problem in the present study.With GPS navigation technology and the popularity of smart mobile devices,you can get a large number of users moving data information from these sources,based on these information can make a lot of meaningful research.Among them,the travel pattern recognition study of trajectory information has become a hotspot in the research of the trajectory data.In the present research method,the trajectory of the feature extraction of confined to basic attributes(speed,acceleration,Angle,etc.).Is proposed in this paper based on the use of permutation entropy as a trajectory classification characteristic values involved in a study,permutation entropy as a measure the properties of time series complexity,this article uses the Angle permutation entropy and Velocity permutation entropy as the characteristics of the track to participate in the travel pattern classification.Track at first,this paper studies permutation entropy and classification accuracy,through the experiment of the same data set,the classification results poor permutation entropy is relatively large.In speed,on the basis of basic properties,such as permutation entropy attribute classification,can improve the accuracy of experimental results.In terms of the choice of classification model,based on the current use of classification model(SVM,Decision Trees,etc)all belong to shallow classification model,this paper adopts the method of Neural Network for depth(Deep Neural Network)model,compared with shallow classification model,the model can use implicit multi-layer complex structure in the Deep layer and the nonlinear transformation,to express the data highly abstract,so as to achieve better classification results,in the use of permutation entropy attribute,on the basis of using Deep Neural Network classification model can further improve the experimental accuracy.In this paper,on the basis of theoretical study,travel pattern classification system was designed and implemented to verify the proposed feature extraction method in this paper and the effectiveness of the classification model.This system adopting a certain number of training sets,on the basis of supervised learning,through the original GPS trajectory data analysis of the path of travel mode,including cars,buses,walking,cycling and the path of the train.Is validated by the experiment,through the classification system of the presented method is feasible and effective.Experimental results show that the extracted by the property and the use of classification method is compared with other methods of classification results accuracy improved.