Research And Application Of Urban Human Trajectory Generation Models

As China’s urbanization process continues to advance,urban planning and management face increasingly severe challenges.The importance of understanding the complex mobility patterns of urban populations is constantly increasing.Urban Human Trajectory Generation is an important means of understanding the mobility patterns of urban populations.This technology extracts human movement patterns from a large amount of trajectory data and generates synthetic trajectories that match the real trajectory movement patterns.Urban Human Trajectory Generation has significant applications in the field of what-if analysis.By generating crowd movement trajectories under specified conditions,it can provide guidance for policy formulation in urban construction,transportation planning,public health,and other fields.On the other hand,Urban Human Trajectory Generation can also be applied to trajectory data de-identification,where personal information is obscured while preserving the crowd movement patterns in the original dataset.However,there are still many issues with current Urban Human Trajectory Generation methods.The complex dependencies between human movement patterns,spatiotemporal contextual information,and real-world environmental constraints in cities make it challenging to model them accurately.At the same time,the technology faces serious data scarcity issues.Due to privacy protection needs,many cities lack high-quality trajectory data,which hinders the promotion and application of Urban Human Trajectory Generation methods.This thesis focuses on researching and applying Urban Human Trajectory Generation methods based on deep learning methods.The main work and innovations are as follows:First,to address the problem that existing methods for generating urban human trajectories cannot effectively learn the complex dependencies between trajectories and urban contextual information,this thesis proposes an Urban Human Trajectory Generation method called MPATG(Mobility Patterns Aware Trajectory Generator)based on Generative Adversarial Networks(GANs).The method uses the discriminator in GAN to learn the crowd movement patterns under different time conditions and convey the pattern information to the generator used for trajectory generation through adversarial training,thereby generating trajectories under different time conditions.In order to introduce real-world constraints into the trajectory generation process,MPATG defines the generation process as a Markov Decision Process(MDP),and defines candidate trajectory points that do not satisfy real-world constraints as invalid actions to avoid them being selected.Finally,this thesis verifies the effectiveness of MPATG using a publicly available GPS trajectory dataset.Second,to address the problem of the lack of trajectory data in most cities,this thesis proposes a transferable Urban Human Trajectory Generation method called T-MPATG(Transfer-Mobility Patterns Aware Trajectory Generator).The model describes human movement patterns as action behaviors driven by movement intentions,enabling the discovery of human movement patterns in semantic environments.Leveraging the similarity of these patterns across different cities,T-MPATG can perform trajectory generation tasks in cities with sparse or even no trajectory data.To address the problem of inconsistent distribution of urban contextual information in different cities,this thesis uses the Transfer Component Analysis algorithm(TCA)to map the city contextual data from different cities into the same latent space,thereby ensuring its consistent distribution.Finally,this thesis verifies the generating performance of T-MPATG under no data and limited data conditions on two publicly available GPS datasets,demonstrating its effectiveness.Third,based on the proposed city population trajectory generation method,this thesis builds an Urban Human Trajectory Generation platform that allows users to specify various generation conditions,call the generation model,and visualize the generated trajectories.The model can also be integrated with other application algorithms to enable the application of city population trajectory generation methods in various downstream tasks.