LSTM trajectory prediction method combining space-time factors and based on graph neural network

Abstract

The invention relates to the technical field of space-time trajectory prediction, in particular to an LSTM trajectory prediction method combining space-time factors and based on a graph neural network, and the method comprises the steps: obtaining the sign-in data of users; performing data preprocessing: performing screening according to the average sign-in number of the users and the social relation condition; dividing a position domain, and endowing the users in the same position domain with the same position label by adopting a clustering method; introducing space-time factors into a gating mechanism, adopting a long-short term memory (LSTM) neural network and a historical trajectory sequence to learn movement habits of a user, and establishing a personal movement trajectory model; inputting the time sequence into a personal movement trajectory model to predict a travel trajectory of the user in a certain time period in the future; and inferring a position semantic category in the travel trajectory prediction result by adopting a position semantic inferring method based on the graph neural network. According to the method, position discovery and extraction are carried out in combination with text content for sparsity of sign-in data generated by a social network based on positions, the number of the positions is expanded, and the prediction accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of spatiotemporal trajectory prediction, in particular to an LSTM trajectory prediction method combining spatiotemporal factors and a graph neural network. Background technique [0002] With the rapid development of the mobile Internet, more and more people begin to actively share their spatio-temporal information on mobile devices through social platforms, usually accompanied by semantic information such as text and pictures that reflect the content of user activities, such as Gowalla and Facebook abroad. , Foursquare and domestic Weibo, circle of friends, etc. Compared with traditional social networks, the biggest difference of location-based social network (LBSN) is that it combines positioning technology, increases the spatial dimension, and is the fusion of location and social interaction. Due to the large number of users and not being constrained by time and place, a large amount of individual movemen...

AI Technical Summary

Problems solved by technology

For example, the time interval between the last check-in and the next check-in of the same user is too long, resulting in missing time series; or the number of check-ins in a certain place is very small, resulting in missing spatial series, and the data sparsity is too strong

In addition, the randomness of user sign-in is relatively large, which is affected by their own attributes such as age and gender and natural factors such as weather.

In the time series, only relying on the check-in data will have a lot of missing, and it is impossible to achieve a good prediction effect

[0006] (2) The construction of the trajectory prediction model is closely related to the historical trajectory prediction. The location where the user appears next time may be related to the previous location, and may be related to several previous locations, which cannot be accurately quantified. Therefore, it is necessary to examine as long as possible the sequence dependence relationship, the current method has the problem of insufficient dependence on long sequences, and it is prone to gradient disappearance, which will affect the accuracy of prediction

Images