Research On Taxi Trajectory Prediction Based On Hidden Markov Model

As a tool of urban public transportation,taxi plays an important role in the daily travel of urban residents.The accurate prediction of taxi trajectory is helpful to the perception and prediction of traffic condition of urban road network and the location-based traffic information service.In order to improve the accuracy of taxi trajectory prediction,this paper proposes a road segment clustering method based on Transformer model to form a hierarchical representation of the road network.Then,the hierarchical hidden Markov model is used to model the spatial hierarchical relationship in the trajectory sequence.At the same time,based on the relationship between spatial levels in the road network,a spatio-temporal relationship fusion network is proposed to express the association between the local road network and the global road network,and the taxi trajectory under different spatial levels in the road network is predicted.The main research contents are as follows:(1)A road network clustering method based on Transformer model is proposed.Firstly,the one hot code is used to embed the physical attributes of road segments.Secondly,considering the global attention among road segment embedded attributes,a K-mean clustering algorithm is constructed based on Transformer model to extract cluster of road segments and the membership relationship between cluster of road segments and road segment.The trajectory sequence represented by the road segment number and the trajectory sequence represented by the cluster of road segments are formed.At the same time,in order to accurately represent the position of vehicles in the road segment,the road segment is further divided into sub-segments to form the sub-segment representation of the trajectory sequence.(2)A multi-level trajectory prediction method based on hierarchical hidden Markov model is proposed.Firstly,the hidden Markov model between the level of cluster of road segments and the level of road segment number,and the hidden Markov model between the level of road segment number and the level of sub-segment are constructed.Secondly,based on GCN,Conv2d and Self-attention,a hierarchical fusion network is constructed to fuse the departure relationship between the cluster of road segments and the distance relationship between the cluster of road segments,as well as the topological relationship of road segments and the departure relationship of road segments under the level of road segment number.The hidden Markov model between cluster of road segments level and road segment number level,and between road segment number level and sub-segment level are optimized.The optimized hierarchical hidden Markov model is used to predict the trajectory hierarchically.In this paper,the proposed method is verified by experiments with the taxi trajectory data of Xi’an city.The experimental results show that the proposed method is superior to the common trajectory prediction models.