Regional rasterization and time fragmentation fused long space-time trajectory prediction method

Abstract

The invention belongs to the technical field of space-time trajectory modeling and feature extraction. The method comprises the steps that 1, a section of continuous space-time trajectory and a targettime point are processed into a coarse-grained space-time trajectory through time fragmentation and region rasterization, and the coarse-grained space-time trajectory is expressed as a fine-grained space-time trajectory through time vectorization and space vectorization; step 2, extracting spatial-temporal features of the coarse-grained spatial-temporal trajectory; 3, space-time characteristics of the fine-grained space-time trajectory are extracted; and step 4, integrating the characteristics of the coarse-grained space-time trajectory and the fine-grained space-time trajectory to obtain predicted trajectory points. According to the prediction method, continuous long space-time trajectory time fragmentation and regional rasterization are carried out, and then unknown trajectory points are predicted through a coarse-grained fragmentation grid convolution module, a fine-grained space-time trajectory prediction module and a coarse-fine joint prediction module.

Description

technical field [0001] The invention relates to the technical field of spatio-temporal trajectory modeling and feature extraction, in particular to a long spatio-temporal trajectory prediction method based on the fusion of regional rasterization and time fragmentation. Background technique [0002] In recent years, emerging mobile applications and services for travel planning and daily life, represented by Didi Taxi, Meituan Waimai, and Baidu Maps, have become an indispensable part of our lives. Predict their travel behavior to provide users with more personalized services. However, the complexity of users' spatio-temporal trajectories can seriously affect the accuracy of their future location predictions. For example, people commute between offices and homes at a relatively fixed time on weekdays, while on weekends, people have a wider range of choices and travel time is more random. Thus, trajectories are more predictable for more regular weekdays. [0003] In addition,...

AI Technical Summary

Problems solved by technology

[0014]ARMIN did not effectively use the spatio-temporal information in the spatio-temporal trajectory data to read and write external storage, so it was unable to select more important spatio-temporal data from the stored long-term historical data Information, insufficient ability in long-term space-time trajectory modeling tasks

Images