The Research And Implementation Of An Urban Traffic Trajectory Data Mining Method

With the development of the Internet of things and the improvement of city intelligence,it becomes more easily to get urban traffic trajectory data.Trajectory data contains rich knowledge to be mined and therefore trajectory data mining becomes an attractive topic.Urban traffic trajectory data mining plays an important role in optimizing decision making for urban,building smart city and reducing traffic congestion.The research of this paper based on an urban electrombile intelligent control system which the author took part in developing.The system collected urban electrombile trajectory data by using Radio Frequency Identification technology and built history database.After analysing related researches and experiments,this paper improves the existing methods and proposes an data mining approach to analyse the trajectory data of this system.We use the urban electrombile trajectory data to finish the experiments,including trajectory preprocessing,stop point identification,trajectory clustering,important route discovery,next position prediction,destination prediction and important route traffic prediction.This paper discusses several approaches have been used to cluster trajectory data and predict future trajectory,then combine Artificial Neural Network trajectory prediction with trajectory clustering to analyse the urban traffic trajectory data.Finally,this paper completes the experiments by using real-world data and analyses the experiment report.The main work is as follows.1.Proposing an improved trajectory partition algorithm based on TRACLUS algorithm.In order to solve the problem of the TRACLUS algorithm ignoring the time factor in trajectory partition phase,we partition the trajectory at the stop point to make subtrajectory more reasonable.As a result,the trajectory cluster accuracy is improved.2.Proposing an approach of prediction deviation correction.Due to the lack of training data,the neural network may get error prediction results which have large deviation.In this case,we try to recognize the potential error results in advance,then use the trajectory clustering approach to find similar trajectory as the prediction result instead.3.To prove the effectiveness of the trajectory prediction approach,we apply the approach to predict trajectory next position,destination and important route flow by using real-world data as training data.We use the result of trajectory cluster to assist the prediction.In the next position prediction experiment,we compare the prediction accuracy rate and find that the prediction deviation correction approach can reduce the deviation and improve the prediction accuracy.The destination experiment has a high prediction accuracy rate,and the important route flow prediction experiment has a less deviation.The experiment results show that the trajectory prediction approach is effective.