Research On Travel Time Prediction And Privacy Protection Mechanism Based On Big Trajectory Data

Big Trajectory Data has become a basic data resource,which is closely related to people’s daily life.As a relatively common application of Big Trajectory Data,travel time prediction is widely used in almost all traffic aspects,such as navigation,route planning,and traffic monitoring.Accurate travel time prediction can greatly improve citizens’ travel experience.The society and academia hope that more trajectory data can be released.At the same time as the research is showing their strengths,since trajectory data usually contains sensitive information from users,directly publishing trajectory data without processing will bring disastrous.As a result,malicious attack and reasoning will pose a serious threat to individual privacy.Therefore,privacy protection technology that takes into account user privacy protection and data availability is the core content of the research on trajectory data publishing.The specific research work of the thesis mainly includes the following three aspects:(1)The thesis analyzes the shortcomings of the current traditional road-based travel time prediction methods,and points out that the model assumptions are too ideal,which can easily lead to error accumulation,which is harmful to the prediction accuracy.Aiming at overcoming the weakness,the pthesis proposes an end-to-end travel time prediction model SLDeep based on learning.The model first standardizes the trajectory samples and inputs them into the SLDeep network for training,uses a bipolar long-and-short-term memory network to capture the sequence features of the trajectory samples,uses a one-dimensional convolutional network to capture the local features of the trajectory samples,and finally outputs the prediction time through a fully connected network.In terms of experiments,the thesis compares the average absolute error,the root mean square error and the coefficient of determination with the LRD model,STTM model and Deep TTE model through the taxi trajectory dataset in Rome and Chengdu.The experimental results prove that the model proposed in this paper has certain advantages in all three indicators,and effectively improves the accuracy of travel time prediction.(2)In order to further improve privacy protection provided by trajectory data publishing mechanism,the thesis proposes the trajectory data publishing mechanism Travelet.The publishing mechanism consists of two algorithms,namely the trajectory generalization algorithm and the trajectory data publishing algorithm.The trajectory generalization algorithm uses an exponential mechanism to select the generalized trajectory to replace the original trajectory,while the trajectory data publishing algorithm adds noise to the counts that contain the original trajectory in the generalized trajectory and publishes it.The above algorithm steps all meet the differential privacy.In terms of experiments,privacy protection,time cost,and utility of the Travelet publishing mechanism proposed in this article get tested on two datasets: Rome trajectory dataset and Oldenburg simulation trajectory dataset.The experiment proves that the Travelet publishing mechanism is effective.Privacy protection is better than other state-of-the-art publishing mechanism.(3)A Big Trajectory Data application and management system get constructed based on above model and publishing mechanism.Two subsystems of the administrator and the user are designed under it.The pre-trained travel time prediction model is used to realize the prediction function.The generalized trajectory generated by the trajectory data set publishing mechanism is used for display,publication and query,and the Big Trajectory Data application and management system is realized based on the Web application development technology.