Research On Vehicle Trajectory Prediction Model For Urban Road Network

With the massive popularity of global positioning system,vehicles in the city are basically equipped with GPS devices.Vehicles in the daily driving will produce a large number of trajectory data.Through the collection,processing and analysis,trajectory data can be used to vehicle trajectory prediction.Vehicle trajectory prediction is to mining the vehicle’s historical movement information to calculate the vehicle’s future location information and behavior dynamics.It plays an important role in traffic safety and location-based services.Vehicle trajectory prediction is one of the research hotspots in the field of intelligent transportation.A large number of related research results have been published so far,among which the traditional statistical-based trajectory prediction model is simple and easy to train,but the accuracy is not high,the prediction model based on behavioral cognition is interpretable,but only applicable to specific scenarios with clear behavioral intention,the trajectory prediction model based on deep learning has strong learning ability and has high accuracy for nonlinear complex trajectory data,but the model is complex,the training time is long,and the prediction results are not interpretable.To improve the prediction accuracy of the model is a common and urgent problem of vehicle trajectory prediction research.In order to improve the accuracy of vehicle trajectory prediction,this paper conducts a research on the vehicle trajectory prediction problem using urban road networks.The main research methods and contents of this paper are as follows:(1)Map data conversion and trajectory data pre-processing.Converting the map data that the original road section is divided into atomic road sections according to location points.Preprocessing the original GPS sampling point trajectory data that the data is cleaned and segmented.And then the original GPS trajectory points are matched to the atomic road sections by parallelized dynamic weighted map matching algorithm,and the GPS trajectory sequence is converted to the atomic road section trajectory sequence.(2)The trajectory prediction model based on pattern mining of indexed set of trajectory sequences is proposed.Firstly,an index set trajectory sequence pattern mining algorithm is proposed for mining time-continuous trajectory sequence patterns.After that,the trajectory sequence patterns mined by this algorithm are used to construct pattern rules by tail division method.Finally,the trajectory prediction is realized by query matching pattern rules.(3)A hybrid model for vehicle trajectory prediction is proposed.To address the shortcomings of a single trajectory prediction model,the CRITIC(Criteria Importance Though Intercrieria Correlation)weight method is used to combine the trajectory prediction model based on indexed set trajectory sequence pattern mining with the trajectory prediction model based on historical transfer probability and the Word2 Vec and LSTM based trajectory prediction model.(4)The hybrid model of vehicle trajectory prediction is applied to the missing trajectory complementation.In this paper,we use Xi’an taxi trajectory data for model training and experimental testing.Firstly,we test the time performance and mining effect of the indexed set trajectory sequence pattern mining algorithm.The experimental results show that under the same environment,the indexed set trajectory sequence pattern mining algorithm has shorter mining time and more trajectory sequence patterns than Apriori Traj and Traj-Prefix Span,and the mining results can be used for trajectory prediction with higher accuracy.Secondly,this paper designs experiments to compare the prediction accuracy of the hybrid model for vehicle trajectory prediction and three single prediction models.The experimental results show that the hybrid model has higher trajectory prediction accuracy.Finally,this paper designs two scenarios for missing trajectory completion,and proves that the hybrid model for vehicle trajectory prediction has certain practical application value in missing trajectory completion.