Trajectory clustering method based on RPCA and depth attention auto-encoder

Abstract

The invention provides a trajectory clustering method based on RPCA and a depth attention auto-encoder, and relates to the field of trajectory clustering methods, and the method comprises the steps: coding a trajectory into a fixed-length feature vector based on a grid division theory; optimizing the deep attention auto-encoder network model by using a near-end method, a back propagation algorithm and an ADMM to achieve the optimal performance; inputting the grid diagonal feature vector into a DAA network to obtain a low-dimensional trajectory feature vector; and finally, gathering all low-dimensional trajectory feature vectors based on K-means. According to the invention, the experimental result shows that compared with a mainstream feature representation-based trajectory clustering method, the algorithm provided by the invention has better anti-noise performance, so each performance evaluation index is better, and different types of trajectories can be distinguished more accurately.

Description

technical field [0001] The invention relates to the field of trajectory clustering methods, in particular to a trajectory clustering method based on RPCA and depth attention autoencoder. Background technique [0002] With the development of the Internet of Things, communication technology, and the improvement of massive data collection and storage capabilities, spatio-temporal trajectory data is easy to obtain and will be efficiently stored in the trajectory database. Trajectory data has basic information such as time stamps and geographic locations, based on which the rich knowledge contained in it can be mined. Trajectory clustering aims to group similar trajectories into the same group to form different clusters, and the trajectories in different clusters have large differences. Among the research directions in many fields of data mining and analysis, trajectory clustering is widely used in solving practical problems such as urban traffic congestion and exploring the law...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem of decreased clustering accuracy caused by noise in trajectory data, the purpose of the present invention is to provide a trajectory clustering method based on RPCA and depth attention autoencoder. First, the trajectory is encoded into a fixed-length The grid diagonal feature vector; secondly, a robust DAA network model is proposed and trained to achieve the best performance; then, the grid diagonal feature vector is input into the robust DAA network model to obtain a low-dimensional trajectory feature vector; Finally, all low-dimensional trajectory feature vectors are aggregated based on K-means

Images