A Comprehensive Review of Dynamic Diffusion Spatio-Temporal Graph Convolutional Networks for Traffic Forecasting

Computer Science and Engineering

Abstract—An essential component of intelligent transportation systems is short-term traffic prediction. Urban management and traffic scheduling greatly benefit from the ability to plan travel routes and prevent traffic congestion by accurately forecasting short-term traffic trends. However, traffic flow is highlyunpredictable and dynamically changes based on neighboring road conditions, making short-term urban traffic forecasting challenging.This research proposes a model based on a Dynamic Diffusion Spatio-Temporal Graph Convolutional Network (DDSTGCN) to address this issue. The model first integrates a dynamic generation matrix with a static distance matrix to capture real-time traffic conditions. It then applies a diffusion random walk approach to model spatial dependencies among nodes. Furthermore,a convolutional LSTM module is used to capture spatiotemporal dependencies in traffic data and improve prediction accuracy.Experimental results, including ablation studies and comparisons with baseline models, demonstrate that the proposed model outperforms existing approaches across multiple evaluation metrics.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.