Research On Taxi Destination Prediction Based On Deep Learning And Attention Mechanism

With the continuous progress of global positioning technology and network communication technology,as well as the rapid growth and popularization of mobile intelligent devices and applications with positioning function,the quantity of trajectory data available to people is increasing.In the era of big data,it is very necessary to analyze and mine large-scale trajectory data.Therefore,trajectory mining has been paid more and more attention by scholars in the industry.Destination prediction based on trajectory data is an important research direction of trajectory mining.However,there are still several major problems in the existing research on destination prediction based on vehicle trajectory.First,due to the constraints of objective conditions,such as acquisition equipment,transmission signals,privacy issues,etc.,in most cases,only a sparse trajectory data set can be obtained,and the number of available moving trajectories is limited and not enough to cover all possible query trajectories,leading to data sparseness problems.The second is that the dependency learning on the long trajectory sequence is not sufficient,forming a long-term dependency problem,less consideration is given to the contextual information of the trajectory.Third,most studies treat every point in the trajectory equally,without emphasizing the key points in the trajectory.To solve these problems,this paper proposes a combination model combining deep learning and attention mechanism to predict the destination of taxi trajectory.The main content of this paper includes: First of all,the taxi trajectory data is preprocessed,and the representation of the taxi trajectory is converted by the grid division method,and the longitude and latitude coordinates of the trajectory points are replaced with a grid.Then,the word embedding technology in natural language processing is introduced to carry out the graphical operation on the taxi trajectory data,and the trajectory data and its metadata are embedded into two-dimensional space to achieve feature embedding.Finally,a destination prediction model based on deep learning and attention mechanism is constructed,by using convolutional neural network and bidirectional long and short-term memory neural network,combined with attention mechanism,to carry out feature learning and training on taxi trajectory data,and then predict the destination.This paper is verified on the real taxi trajectory data set in Kunming City.The experimental results show that the model proposed in this paper performs better than other models in terms of prediction.