Saved in:
Bibliographic Details
Main Authors: Fang, Jiayu, Shao, Zhiqi, Choy, S T Boris, Gao, Junbin
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.13433
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915451176484864
author Fang, Jiayu
Shao, Zhiqi
Choy, S T Boris
Gao, Junbin
author_facet Fang, Jiayu
Shao, Zhiqi
Choy, S T Boris
Gao, Junbin
contents Spatio-temporal traffic forecasting is challenging due to complex temporal patterns, dynamic spatial structures, and diverse input formats. Although Transformer-based models offer strong global modeling, they often struggle with rigid temporal encoding and weak space-time fusion. We propose STPFormer, a Spatio-Temporal Pattern-Aware Transformer that achieves state-of-the-art performance via unified and interpretable representation learning. It integrates four modules: Temporal Position Aggregator (TPA) for pattern-aware temporal encoding, Spatial Sequence Aggregator (SSA) for sequential spatial learning, Spatial-Temporal Graph Matching (STGM) for cross-domain alignment, and an Attention Mixer for multi-scale fusion. Experiments on five real-world datasets show that STPFormer consistently sets new SOTA results, with ablation and visualizations confirming its effectiveness and generalizability.
format Preprint
id arxiv_https___arxiv_org_abs_2508_13433
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle STPFormer: A State-of-the-Art Pattern-Aware Spatio-Temporal Transformer for Traffic Forecasting
Fang, Jiayu
Shao, Zhiqi
Choy, S T Boris
Gao, Junbin
Artificial Intelligence
Spatio-temporal traffic forecasting is challenging due to complex temporal patterns, dynamic spatial structures, and diverse input formats. Although Transformer-based models offer strong global modeling, they often struggle with rigid temporal encoding and weak space-time fusion. We propose STPFormer, a Spatio-Temporal Pattern-Aware Transformer that achieves state-of-the-art performance via unified and interpretable representation learning. It integrates four modules: Temporal Position Aggregator (TPA) for pattern-aware temporal encoding, Spatial Sequence Aggregator (SSA) for sequential spatial learning, Spatial-Temporal Graph Matching (STGM) for cross-domain alignment, and an Attention Mixer for multi-scale fusion. Experiments on five real-world datasets show that STPFormer consistently sets new SOTA results, with ablation and visualizations confirming its effectiveness and generalizability.
title STPFormer: A State-of-the-Art Pattern-Aware Spatio-Temporal Transformer for Traffic Forecasting
topic Artificial Intelligence
url https://arxiv.org/abs/2508.13433