Research On Taxi Trajectory Data Recovery And Anomaly Detection For Intelligent Transportation

In recent years,with the rapid development of urbanization,cities have been expanding in size,and people’s daily travel increasingly relies on public transportation.Public transportation,represented by taxis and ride-hailing services,has become an essential means of transportation for people’s daily commuting.Taxi trajectory data serves as an important data source for optimizing taxi services,traffic management,and urban road planning.Researching taxi trajectory data can provide real-time vehicle dispatching and route planning for taxi service companies,optimizing taxi services.It can also offer real-time monitoring and prediction of traffic conditions for urban traffic management departments,assisting urban planning departments in better road network planning,traffic guidance,and congestion control.Taxis in cities are equipped with GPS.These GPS devices collect latitude and longitude information as well as passenger occupancy status at a certain frequency and upload them to cloud servers in real-time.However,due to weak GPS signals,congested transmission networks,and faulty receiving devices,the GPS monitoring system may fail to provide reliable positioning accuracy,resulting in inevitable data gaps and errors during data collection and transmission.Analyzing and mining research using incomplete and erroneous taxi trajectory data can lead to incorrect results,necessitating the completion of missing trajectory data.Furthermore,these taxis and ride-hailing services,while providing convenience to urban residents,inevitably give rise to certain issues.Mainly,some taxi drivers deliberately take longer routes to earn more profit,while some ride-hailing drivers violate regulations by taking shortcuts to save time.To safeguard the legitimate rights and interests of the people and strengthen the supervision of taxis and ride-hailing services,it is necessary to conduct research on anomaly detection in taxi passenger trajectory data.Currently,there have been some models and algorithms proposed for the recovery of missing trajectories and the detection of abnormal passengers’ trajectory data.However,due to the variable sampling frequency of trajectory data and the complexity of travel routes,existing models have not effectively addressed these challenges.In light of these issues,this study proposes a model for trajectory completion based on trajectory similarity and dynamic mask training.It fully explores the similarity features between missing trajectories and historical trajectories and improves the accuracy of trajectory completion by using the dynamic mask training method to increase the amount of training data.For the detection of anomalies in passenger trajectory data,an anomaly detection algorithm based on the DTW(Dynamic Time Warping)similarity centroid average trajectory is proposed.This algorithm addresses the issue of varying trajectory lengths caused by the variable sampling frequency,thereby improving the accuracy and real-time performance of anomaly detection.The main research contents of this dissertation are as follows:(1)In response to the problem of sparse trajectories and the limitations of existing trajectory completion models in effectively extracting multi-level similarity features between historical trajectories,this dissertation proposes the trajectory completion model called Similar Move,based on sequence similarity and dynamic mask training.This model utilizes a gated graph neural network to learn grid embedding vector representations.It leverages the DTW similarity and attention mechanism to extract multi-level similarity features from historical trajectories.Additionally,it utilizes self-attention mechanism to learn relevant features of the missing trajectories,effectively addressing the issue of extracting multi-level similarity features between historical trajectories.Lastly,the model incorporates the dynamic mask training method during the training process to enhance the inclusion of location information,effectively alleviating the problem of sparse location information.(2)In light of the issues regarding real-time performance and accuracy in traditional anomaly trajectory detection algorithms based on trajectory similarity,this dissertation proposes an anomaly detection algorithm based on DTW similarity and centroid average trajectory.Firstly,to address the problem of sequence misalignment caused by different sampling frequencies in taxi trajectories,an improved DTW algorithm called FastDTW is employed to calculate the similarity between trajectory sequences.Next,a trajectory voting method based on DTW similarity is proposed to efficiently detect a portion of normal trajectories by controlling the number of trajectory sequences.Lastly,to overcome the low real-time performance of most anomaly trajectory detection methods for static trajectories,an anomaly trajectory detection algorithm based on centroid average trajectory is introduced.It continuously updates the "average trajectory" during the detection process and outputs real-time detection results,thereby improving the real-time performance of the detection process.(3)This dissertation conducted trajectory completion experiments on taxi datasets from three cities: Beijing,Shenzhen,and Chengdu.The experimental results demonstrate that the proposed trajectory recovery model outperforms existing models in terms of recovery recall and average error.For anomaly trajectory detection,experiments were conducted on the taxi datasets from Chengdu and Los Angeles.The results show that the proposed anomaly detection algorithm performs better than existing algorithms in terms of the F1 score.Furthermore,this dissertation conducted joint analysis experiments on trajectory anomaly detection and recovery using the datasets from Chengdu and Los Angeles.The results indicate that using the Similar Move model proposed in this dissertation for trajectory recovery can improve the F1 score of anomaly detection.