| Traffic trajectory data is the data that describes the real-time location information of moving objects over time.It can provide many information,including travel mode,travel law,traffic state distribution,causes of traffic congestion and so on.Due to the long-term and multi-dimensional characteristics of track data,the existing traffic big data analysis methods are difficult to accurately and comprehensively describe the track characteristics,identify the travel mode of track objects and mine potential travel modes.To solve the above problems,this paper focuses on traffic travel mode recognition and traffic travel mode analysis based on GPS trajectory data.The main research work is as follows:(1)In view of the large difference of GPS track data and the fact that the similarity measurement based on Euclidean distance can not meet the requirements of track similarity measurement,a travel mode recognition algorithm based on a two-stage manifold learning model is proposed.Firstly,the spatiotemporal features of GPS trajectory are extracted and expressed as a symmetric positive definite(SPD)matrix.Using the similarity measure of SPD manifolds and Riemannian manifolds,a global cost function is constructed by accumulating the distance between all training samples,and then minimizing the surrogate function to learn a conversion matrix,so as to identify the travel mode of the test samples.Through numerical experiments,it is verified that the performance of the traffic travel mode recognition algorithm based on two-stage manifold learning proposed in this paper is better than the traditional machine learning algorithm.(2)Aiming at the problems that the traditional mixed travel mode identification methods have no obvious difference in characteristics and low accuracy of detection when describing different travel modes,a mixed travel mode detection algorithm based on ordered subspace clustering on manifold is proposed.The trajectories of mixed travel modes are screened out from the dataset and expressed as a set of symmetric positive definite matrices.The identification of mixed travel modes is realized by combining SPD manifold measurement with low rank representation model.Experiments show that the hybrid travel mode recognition algorithm proposed in this paper has good performance.(3)In view of the large proportion of human subjective components and large differences in data distribution in the traditional travel mode analysis methods,this thesis proposes a travel mode analysis method based on sparse subspace clustering to find the travel modes in the travel trajectory.According to the property that behavior modes(trajectories)can be linearly represented by the approximate linear combination of the same type of patterns,firstly,the trajectory data set is reconstructed,the complete trajectory is divided into sub trajectory segments with uniform length,and the training data is constructed;Then,we use the sparse subspace clustering model to learn a representation matrix.The list of the matrix shows the weight of one sample when representing other samples.The sample with the largest weight can be used as the representative sample of trajectory data to represent the travel modes.The experimental results show that some travel modes inside the Sixth Ring Road of Beijing mainly include the following four types: residential area to office area,residential area to shopping center,residential area to scenic spot and university travel modes.In addition,the travel mode from Weigongcun residential area to Beijing Zoo through the capital gymnasium has the greatest positive impact on the surrounding road sections(that is,the flow change of this road section is directly proportional to the surrounding road sections);The travel mode from Fengtai residential area to Haidian office area through the West Fourth Ring Road has the greatest negative impact on the surrounding road sections(that is,the flow change of this road section is inversely proportional to the surrounding road sections). |
PDF Full Text Request