SiTC-Diff: A Spatial–Temporal Integrated Correction Diffusion model for multi-step traffic flow forecasting

Yuan Xu

doi:10.1016/j.knosys.2026.116914

SiTC-Diff: A Spatial–Temporal Integrated Correction Diffusion model for multi-step traffic flow forecasting

2026-08-27

0

PRE

AI

DOI：10.1016/j.knosys.2026.116914

Original

Original request for help

Save

Abstract

En 中文

• Proposes SiTC-Diff for multi-step traffic flow forecasting. • Introduces CPC to condition reverse innovations on historical traffic context. • Formulates traffic evolution with multi-frame historical conditioning. • Designs WRAUNet with WinTSSA and DBRA for spatiotemporal denoising. • Achieves leading forecasting accuracy on four real-world traffic datasets.

Keywords:

Traffic flow forecasting

Diffusion models

Contextual path correction

Spatiotemporal representation learning

Conditional generation

SiTC-Diff: A Spatial–Temporal Integrated Correction Diffusion model for multi-step traffic flow forecasting

Abstract

Journal

Organization

Cited Papers