Traffic flow prediction method and device based on dynamic ordinary differential graph neural network

Abstract

This invention discloses a traffic flow prediction method and apparatus based on a dynamic ordinary differential graph neural network. The method employs a single encoder architecture, and the system modules include: a module for establishing a road traffic network semantic matrix, a spatial dependency capture module, a temporal dependency extraction module, a dynamic graph ordinary differential network module, a graph convolutional neural network module, a feature fusion gating unit, a traffic flow prediction module, and a terminal design module. To address the over-smoothing problem of node data in traditional traffic flow prediction methods, this invention introduces a dynamic graph ordinary differential network. It uses ordinary differential equations to abstract analytical solutions to the road traffic network topology, extracts spatiotemporal features by constructing a spatiotemporal attention module, and further captures spatiotemporal data features of road traffic flow using a feature fusion gating unit, thereby improving the accuracy of traffic flow prediction.

Images