Research On Vehicle Behavior Prediction Based On Neural Network

Predicting travel trajectory of vehicles can not only provide users with personalized services,but also have a certain effect on traffic guidance,traffic control and urban construction.There are two aspects of trajectory prediction: on the one hand,it is necessary to mine movement pattern from data sets,especially in three-dimensional traffic environment,trajectory data is easy to be confused,and a special model should be built to fit traffic network.On the other hand,digging deep vehicle driving motives from movement pattern is import for predicting next location based on current state.Based on traffic network in complex road environment,this paper considers the semantic association between trajectory nodes,uses high-dimensional traffic network to shield trajectory chaos caused by the sample latitude and longitude information of two-dimensional space,and build an integrated recurrent neural network model using the particularity of trajectory node sequence based on high-dimensional traffic network.The main research work and contributions of the paper are as follows:1.Aiming at semantic association between trajectory nodes during driving process,vehicle driving motion pattern is extracted,we propose a traffic network construction method based on high dimensional vector.Firstly,window mechanism is used to extract semantic associations inside the associated trajectory nodes.Based on this,the trajectory node corpus is constructed,so that the context relationship between trajectory nodes is contained in the node corpus.Secondly,for the context relationship of each trajectory node,we use embedded network to embed semantic association between trajectory nodes into high-dimensional vector,making it a measurable probability distribution form,thus constructing a traffic network space in high-dimensional space.Trajectory confusion problem caused by latitude and longitude limitation is shielded.2.We analysis the temporal correlation between trajectory nodes deeply,and use integrated recurrent neural network to predict trajectory,this method can accurately and objectively predict the next position vehicle will arrive.Firstly,different length trajectories are used to perform fitting learning with different variants of recurrent neural network,and different sub-learners are used to train trajectory multi-level features.Then,using bagging method to integrate output results of each learner to improve accuracy of trajectory prediction.Finally,experimental verification is carried out through real trajectory dataset of a certain city in China.Experiments show that the high-dimensional traffic network construction algorithm used in this paper can not only reflect real road vehicle trajectory characteristics,but also play a role in some applications;the integrated recurrent neural network is compared with traditional probability statistical model and single neural network work model.The integrated network model predicts better results and better reflects the true trajectory of vehicle.