Research On Transportation Mode Recognition Methods Based On GPS Data

China's economic construction and industrial development greatly promote the process of urbanization,and at the same time,put forward higher requirements for the city's management ability.The expansion of urban territory and the growth of permanent resident population bring more motor vehicles,which lead to severe challenges such as traffic congestion and environmental pollution for urban traffic system.The travel information of residents determines the demand of urban transportation.The information of residents' travel behavior can be summarized and analyzed,and the characteristics and rules of residents' travel can be studied,which can provide decision support for the important application of urban traffic system such as urban planning,road layout and traffic law formulation.The traditional way to obtain the information of residents' travel behavior is mainly based on the method of manual collection,and the data quality cannot be guaranteed.In recent years,with the improvement and popularization of GPS devices,the large scale of GPS data produced provides a new solution for the acquisition of travel behavior information.Currently,some studies rely on data other than GPS to help with recognition,such as accelerometer and gyroscope sensor data or GIS information data.However,these data are not always available in actual situation;At the same time,some studies first find the change point through the label file,and then divide the trajectory into several segments of single transportation mode,which is a disguised simplification of the task of transportation mode recognition.Based on the raw GPS data,this paper proposes a trajectory segmentation method with fixed length threshold,constructs a 72-dimensional feature vector containing global feature and point feature,and designs a transportation mode recognition model based on ensemble learning.The research contents of this paper mainly include:(1)Feature extraction based on GPS data.In this paper,the trajectory segmentation method based on fixed length rule is adopted,which is more in line with the actual situation.In addition to the regular modes of transportation(walking,biking,bus,driving,subway,and train),the final category includes the case that a certain segment contains multiple transportation modes,known as hybrid modes,to more accurately determine when and where residents change their travel behavior.(2)Discussion of the classification ability of the ensemble learning model represented by Deep Forest in transportation mode recognition task.The optimization method of Deep Forest is studied,and four individual classifiers,including Random Forest,completely Random Forest,Support Vector Machine and XGBoost,are integrated to improve the generalization ability and robustness of the final ensemble model.The experimental results verify the effectiveness of the proposed method.The improved Deep Forest has achieved the results with an accuracy of 88.6%,which is betterthan all the comparison experiment models that we set,including Random Forest,XGBoost and Convolution Neural Network.