Research On The Prediction Of Vehicle’s Next Location Based On Road Network

With the continuous development of Global Navigation Satellite System(GNSS)and wireless sensing technology,as well as the increasing popularity of mobile positioning devices and surveillance cameras,people are increasingly benefiting from various Location Based Services(LBS),such as entertainment services,advertising,parking space information and traffic congestion.The core task of these services is to collect the location information of users or environmental objects,and analyze and predict the location of the next moment based on the current trip in advance,so as to provide users with more timely and accurate services.The next moment position prediction problem for moving vehicles has been widely studied.The existing methods can be roughly divided into two categories according to the types of trajectory generation.The first is to realize the next position prediction through the location data generated at the vehicle end or the vehicle traffic data collected at the road monitoring bayonet.Second,it focuses on the prediction of continuous Point of Interest(POI)and POI recommendation with check-in data on social networks,so it does not consider the constraints related to road information in the traffic system.Because of the different characteristics of the two types of trajectory data,each position prediction method can not be well applied to both.Therefore,this paper focuses on the discussion of the first type of prediction method,that is,using the traffic track data of vehicles to predict the position of vehicles at the next moment,and correspondingly puts forward the research on the position prediction of vehicles at the next moment facing the road network.The main work is described as follows:1.Research on the prediction of the vehicle’s next moment position based on trajectory start point fusion and similar trajectory clustering.In view of the shortcomings of the traditional Markov model,which simply considers the position just passed to predict the next position,but ignores the position passed earlier in the trajectory,which is difficult to make full use of the existing information to predict,this paper proposes a travel time difference model integrating the starting point.The model takes into account that users tend to choose the shortest route from origin to destination,and predicts the next location by comparing the difference between the shortest drive time from the location passed to the next location and the actual travel time.In addition,in order to solve the problem that the traditional Markov model does not care about user preference when predicting,that is,as long as the user is in the same current position,the prediction will point to the same next position,this paper also proposes a Markov model based on similar trajectory clustering.The model takes into account the consistency of the trajectories between users,and a Markov model can be trained by the trajectories of all users.Finally,two single models are used for joint prediction and the scheme performance is verified by simulation.The simulation results show that compared with the prediction of a single model,the proposed joint model improves the accuracy of predicting the next position by 59.6% compared with the traditional Markov model.2.Research on the prediction of the vehicle’s next moment position based on multiple features embedded in LSTM.The single prediction model proposed in research point 1 only considers the influence of driving time factor and different users’ trajectory sequence on location prediction respectively.Due to the limitation of statistical probability model itself,it is difficult to combine user’s personal interest preference and space-time characteristics of moving state into the same model for prediction.Therefore,this paper proposes to first map the user trajectory data into the low-dimensional potential space using embedding technology to get the feature vector.Secondly,the use of Long Short-Term Memory network and attention mechanism as the subsequent learning model can more effectively mine the potential features of user trajectory data,so as to more accurately identify the hidden movement patterns of users.Finally,the scheme performance is tested and verified by simulation.Simulation results show that compared with the joint model proposed in Point 1 and other related models,the proposed method based on multiple features embedding LSTM has higher prediction accuracy.