Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Traffic Forecasting against Spatial Shift over Years

Hongjun Wang, Jiyuan Chen, Tong Pan, Zheng Dong, Lingyu Zhang, Renhe Jiang, Xuan Song

TL;DR

This paper investigates state-of-the-art models using newly proposed traffic OOD benchmarks and finds that these models experience a significant decline in performance, and proposes a novel Mixture of Experts (MoE) framework, which can be seamlessly integrated into any spatiotemporal model, outperforming current state-of-the-art approaches in addressing spatial dynamics.

Abstract

Recent advancements in Spatiotemporal Graph Neural Networks (ST-GNNs) and Transformers have demonstrated promising potential for traffic forecasting by effectively capturing both temporal and spatial correlations. The generalization ability of spatiotemporal models has received considerable attention in recent scholarly discourse. However, no substantive datasets specifically addressing traffic out-of-distribution (OOD) scenarios have been proposed. Existing ST-OOD methods are either constrained to testing on extant data or necessitate manual modifications to the dataset. Consequently, the generalization capacity of current spatiotemporal models in OOD scenarios remains largely underexplored. In this paper, we investigate state-of-the-art models using newly proposed traffic OOD benchmarks and, surprisingly, find that these models experience a significant decline in performance. Through meticulous analysis, we attribute this decline to the models' inability to adapt to previously unobserved spatial relationships. To address this challenge, we propose a novel Mixture of Experts (MoE) framework, which learns a set of graph generators (i.e., graphons) during training and adaptively combines them to generate new graphs based on novel environmental conditions to handle spatial distribution shifts during testing. We further extend this concept to the Transformer architecture, achieving substantial improvements. Our method is both parsimonious and efficacious, and can be seamlessly integrated into any spatiotemporal model, outperforming current state-of-the-art approaches in addressing spatial dynamics.

Robust Traffic Forecasting against Spatial Shift over Years

TL;DR

This paper investigates state-of-the-art models using newly proposed traffic OOD benchmarks and finds that these models experience a significant decline in performance, and proposes a novel Mixture of Experts (MoE) framework, which can be seamlessly integrated into any spatiotemporal model, outperforming current state-of-the-art approaches in addressing spatial dynamics.

Abstract

Recent advancements in Spatiotemporal Graph Neural Networks (ST-GNNs) and Transformers have demonstrated promising potential for traffic forecasting by effectively capturing both temporal and spatial correlations. The generalization ability of spatiotemporal models has received considerable attention in recent scholarly discourse. However, no substantive datasets specifically addressing traffic out-of-distribution (OOD) scenarios have been proposed. Existing ST-OOD methods are either constrained to testing on extant data or necessitate manual modifications to the dataset. Consequently, the generalization capacity of current spatiotemporal models in OOD scenarios remains largely underexplored. In this paper, we investigate state-of-the-art models using newly proposed traffic OOD benchmarks and, surprisingly, find that these models experience a significant decline in performance. Through meticulous analysis, we attribute this decline to the models' inability to adapt to previously unobserved spatial relationships. To address this challenge, we propose a novel Mixture of Experts (MoE) framework, which learns a set of graph generators (i.e., graphons) during training and adaptively combines them to generate new graphs based on novel environmental conditions to handle spatial distribution shifts during testing. We further extend this concept to the Transformer architecture, achieving substantial improvements. Our method is both parsimonious and efficacious, and can be seamlessly integrated into any spatiotemporal model, outperforming current state-of-the-art approaches in addressing spatial dynamics.
Paper Structure (8 sections, 7 equations, 4 figures, 5 tables)

This paper contains 8 sections, 7 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traffic Forecasting
  4. OOD in Spatio-Temporal Analysis
  5. Problem Statements
  6. Methodology
  7. Experiment
  8. Conclusion

Figures (4)

  • Figure 1: In (a) and (b), we compare the test performance of mainstream ST-GNNs: GWNet wu2019graph, AGCRN bai2020adaptive, MTGNN wu2020connecting, TrendGCN jiang2023enhancing and STAEformer liu2023spatio on in- and out-of-distribution, respectively. The statistical results for PEMS03 and PEMS04 are shown in (c) and (d), respectively. For each dataset, we illustrate the distribution of Kendall’s $\tau$kendall1938new and DTW muller2007dynamic, which present the similarity of graph relations and temporal distributions.
  • Figure 2: The schematic of targeted graphons generation in training and testing period.
  • Figure 3: Our method integrates seamlessly with conventional ST-GNNs by incorporating a learnable expert graphons layer. The left (grey) portion of the figure represents the traditional ST-GNNs pipeline, which we enhance by replacing the standard learnable graph construction module with our expert graphons framework. Both approaches serve the core function of preparing the adjacency matrix for the spatiotemporal module, ensuring compatibility and easy integration.
  • Figure 4: Searching for the optimal number of divisions is depicted for each dataset in the upper figures. The corresponding performance of GWNET, TrendGCN, AGCRN, and MTGNN under different numbers of divisions for each dataset is illustrated at the bottom of the figure.

Theorems & Definitions (1)

  • Definition 4.1: Maximum Spatiotemporal Graph Division (MSGD)