Research On K Nearest Neighbor Query Of Trajectory

The application of location information generates a large volume of space-time data with multidimensional attributes such as latitude,longitude and time.The traditional index-based k nearest neighbor track query method can’t obtain the space-time semantic features of the track,and its track representation mode ignores the space-time correlation of track points and has low efficiency in large-scale track data query.Although people have carried out extensive research in the field of trajectory data mining,there has not been a set of mature solutions for how to transform different types of trajectories into effective representation to make it more conducive to data mining and other operations.Aiming at the problem that traditional trajectory data representation method is difficult to get the space-time correlation of the trajectory,this paper put forward a trajectory encoder based on the sparse multi-head self-attention mechanism,and uses the trajectory encoder to extract the semantic features of the trajectory data to obtain the encoding vector of the trajectory.The obtained trajectory coding vector is more effective than the previous trajectory representation methods such as Geohash index and R tree index.Aiming at the low efficiency of trajectory k nearest neighbor query method,a new method based on trajectory coding vector and local sensitive hash was proposed to adapt to large-scale trajectory data query.To solve the problem of low efficiency and poor accuracy of trajectory reverse k nearest neighbor query,a neural network named TFPNN was constructed based on the cross of trajectory features.TFPNN was used to extract high-level cross semantic features between different trajectory data.Compared with the previous track representation methods such as Geohash index and R tree index,the proposed cross-semantic coding vector of track features is more effective.The experimental results demonstrate that the proposed method is better than the current query methods in terms of accuracy and efficiency,and can be used in the actual road network trajectory data set.