Knowledge-Enhanced Graph Neural Network for Traffic Flow Prediction

Abstract

To address the challenges of existing traffic volume prediction methods in effectively integrating heterogeneous data and mining deep spatio-temporal semantic correlations, this paper proposes a prediction method based on a Spatio-Temporal Knowledge Graph (STKG). The approach first constructs a knowledge graph to uniformly model road network topology, dynamic traffic flow, and external influencing factors. Then, a knowledge-enhanced spatio-temporal graph neural network is designed to deeply integrate knowledge semantics with real-time data through a unique fusion mechanism, aiming to capture the complex evolutionary patterns of the traffic system. Experimental results demonstrate that the method significantly enhances prediction accuracy in complex road network and sudden event scenarios, improving semantic association accuracy by 17.4% and reducing the Root Mean Square Error by 15.2%.