The Technology Research On Spatio-temporal Feature Mining Of Private Car Trajectory Data

With the rapid development of economy and society and the rapid progress of urbanization in China,the transportation infrastructure has been greatly improved.Private cars have become the transportation needs of the people for a better life.Citizens driving private cars has become an important way to travel.At the same time,with the popularization of network and the rapid development of communication technology,a variety of vehicle intelligent sensing devices are widely used,such as intelligent car machine,cloud rearview mirror,OBD box and so on.Through these intelligent sensing devices,a large number of spatio-temporal data such as vehicle trajectory can be obtained,which makes it possible to perceive the travel information of citizens.In the era of big data,it is of great economic value to carry out the research on the trajectory data of private cars and mine the trajectory data deeply,which is of great significance for the government and transportation management departments to make scientific decisions on intelligent transportation construction.However,in trajectory data mining,due to the characteristics of trajectory data such as spatio-temporal sequence,sparsity,noise interference and deep-rooted spatio-temporal attributes,trajectory data mining is different from traditional data mining.There are many technical problems that need to be studied,such as data imbalance,data sparsity and so on.In view of the technical difficulties to be solved in current trajectory data mining,this thesis studies the four aspects of trajectory data compression,spatio-temporal attribute mining,time length prediction and spatial prediction of user mobile behavior.This thesis runs through the processes of trajectory data processing,trajectory data mining and intelligent prediction,and focuses on the research progress and technical challenges related to trajectory compression technology,trajectory mining technology and behavior prediction.The main work of this thesis is summarized as follows:(1)In the era of rapid development of information technology,massive trajectory data has brought huge transmission and storage costs,as well as the demand of many locationbased application services for high sampling rate and real-time acquisition of trajectory data.In this thesis,an opportunistic compression method(OCT-LSTM)for large-scale private car trajectory collection process is proposed.Through the prediction model to learn and make use of the spatio-temporal attributes contained in the vehicle trajectory data,the prediction value is used to replace the strategy of the transmission process,and the compression in the process of vehicle trajectory data transmission is realized.Specifically,the trajectory data is first simplified into two parts,spatial path and time-distance sequence,by a spatiotemporal change method,and compressed separately.,and compressed respectively.For the spatial path data,the dictionary compression method is adopted;for the temporal distance sequence,the prediction model based on long-term and short-term memory network is trained to remember and learn the movement patterns in the historical data and to make predictions.OCT-LSTM method can significantly reduce the transmission overhead of trajectory data(approximately 127.3 times compression ratio when compression error is less than 25),and can realize low-latency trajectory compression processing(calculation delay is about 1.81 seconds),which is beneficial for real-time transmission of vehicle trajectory data with low cost and high sampling rate.(2)In the face of the deep spatio-temporal information contained in the private car trajectory data(such as spatial similarity,multi-scale movement and time periodicity,etc.),a user travel behavior mining method based on private car trajectory data is designed.First of all,an on-line detection method of vehicle state based on sliding window is proposed,which can detect the state of vehicle immediately from the trajectory data collected in real time.The abstract user trajectory data of private car is transformed into valuable mobile behavior data.Then,the spatial attributes of private car user mobility behavior data are mined,a density clustering method of user aggregation effect is proposed,and the spatial distribution characteristics and multi-scale mobility attributes of user mobility behavior are analyzed.Finally,the time attribute of user mobile behavior data is mined,and a method to measure the spatio-temporal difference of user travel behavior is given.(3)Stay time is one of the important indicators to understand users' stay behavior and travel motivation.However,the users' behavior is affected by many factors and has the characteristic of randomness.How to effectively use trajectory data to predict the duration of stay behavior of users is still a challenging problem.A stay time prediction method(STP)for private car users' stay behavior is studied in this thesis.In the construction of data-driven intelligent model,a spatio-temporal feature extraction method based on density clustering and kernel density estimation is designed.Targeted use of the private car users' stay behavior data of spatial similarity,time periodicity and spatio-temporal correlation characteristics.Combined with the prediction model of users' stay behavior based on Gradient Boosted Regression Tree,and through the large-scale real private car trajectory data to verify the experimental method,the STP method can achieve the prediction effect of RMSE value of about 123.94(R2 value of about 0.893).(4)In-depth understanding of the mobile behavior of users and accurately predicting the future travel destination of users has become an important basic condition for the realization of location-related intelligent services.Taking advantage of the exploratory and spatio-temporal dependence of user travel behavior,this thesis further explores the location prediction method(TBP)for private car users' travel behavior,analyzes the exploratory travel behavior in the user travel behavior(visiting the new location),and designs a prediction method(TBP-E)for the private car users to explore the travel behavior,which can predict whether the user's travel destination is a new location that has never been visited.Furthermore,this thesis studies the spatio-temporal gravity model of user travel by mining the spatio-temporal association attribute of the user's travel destination,and proposes a location prediction method(TBP-L)for private car users' travel behavior,which can effectively predict the user's travel destination.