Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Temporal Attention Evolutional Graph Convolutional Network for Multivariate Time Series Forecasting

Xinlong Zhao, Liying Zhang, Tianbo Zou, Yan Zhang

TL;DR

This work tackles multivariate time series forecasting with dynamic spatial relationships and interactions across multiple time scales. It introduces TAEGCN, a unified framework that combines Temporal Multi-head Self-Attention (TMSA) for temporal feature extraction with Evolvable Graph Construction (EGC) for learning evolving graph structures, integrated through graph convolutional layers. Key contributions include the design of TMSA and EGC modules, a layerwise spatio-temporal architecture with residual connections, and extensive ablation and sensitivity analyses on METR-LA and PEMS-BAY showing improvements over strong baselines, especially at longer horizons. The approach offers practical benefits for traffic forecasting and other domains with non-stationary inter-variable relationships by adaptively modeling both time-varying dependencies and scale-aware temporal patterns.

Abstract

Multivariate time series forecasting enables the prediction of future states by leveraging historical data, thereby facilitating decision-making processes. Each data node in a multivariate time series encompasses a sequence of multiple dimensions. These nodes exhibit interdependent relationships, forming a graph structure. While existing prediction methods often assume a fixed graph structure, many real-world scenarios involve dynamic graph structures. Moreover, interactions among time series observed at different time scales vary significantly. To enhance prediction accuracy by capturing precise temporal and spatial features, this paper introduces the Temporal Attention Evolutional Graph Convolutional Network (TAEGCN). This novel method not only integrates causal temporal convolution and a multi-head self-attention mechanism to learn temporal features of nodes, but also construct the dynamic graph structure based on these temporal features to keep the consistency of the changing in spatial feature with temporal series. TAEGCN adeptly captures temporal causal relationships and hidden spatial dependencies within the data. Furthermore, TAEGCN incorporates a unified neural network that seamlessly integrates these components to generate final predictions. Experimental results conducted on two public transportation network datasets, METR-LA and PEMS-BAY, demonstrate the superior performance of the proposed model.

Temporal Attention Evolutional Graph Convolutional Network for Multivariate Time Series Forecasting

TL;DR

This work tackles multivariate time series forecasting with dynamic spatial relationships and interactions across multiple time scales. It introduces TAEGCN, a unified framework that combines Temporal Multi-head Self-Attention (TMSA) for temporal feature extraction with Evolvable Graph Construction (EGC) for learning evolving graph structures, integrated through graph convolutional layers. Key contributions include the design of TMSA and EGC modules, a layerwise spatio-temporal architecture with residual connections, and extensive ablation and sensitivity analyses on METR-LA and PEMS-BAY showing improvements over strong baselines, especially at longer horizons. The approach offers practical benefits for traffic forecasting and other domains with non-stationary inter-variable relationships by adaptively modeling both time-varying dependencies and scale-aware temporal patterns.

Abstract

Multivariate time series forecasting enables the prediction of future states by leveraging historical data, thereby facilitating decision-making processes. Each data node in a multivariate time series encompasses a sequence of multiple dimensions. These nodes exhibit interdependent relationships, forming a graph structure. While existing prediction methods often assume a fixed graph structure, many real-world scenarios involve dynamic graph structures. Moreover, interactions among time series observed at different time scales vary significantly. To enhance prediction accuracy by capturing precise temporal and spatial features, this paper introduces the Temporal Attention Evolutional Graph Convolutional Network (TAEGCN). This novel method not only integrates causal temporal convolution and a multi-head self-attention mechanism to learn temporal features of nodes, but also construct the dynamic graph structure based on these temporal features to keep the consistency of the changing in spatial feature with temporal series. TAEGCN adeptly captures temporal causal relationships and hidden spatial dependencies within the data. Furthermore, TAEGCN incorporates a unified neural network that seamlessly integrates these components to generate final predictions. Experimental results conducted on two public transportation network datasets, METR-LA and PEMS-BAY, demonstrate the superior performance of the proposed model.
Paper Structure (13 sections, 6 equations, 7 figures, 3 tables)

This paper contains 13 sections, 6 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Temporal Attention Evolutional Graph Convolutional Network
  4. Temporal Multi-head Self-Attention
  5. Evolvable Graph Construction
  6. Experiments
  7. Setup & Datasets
  8. Baselines
  9. Results
  10. Ablation study
  11. Study of TMSA
  12. Study of EGC
  13. Conclusion

Figures (7)

  • Figure 1: The framework of TAEGCN
  • Figure 2: The framework of TMSA, each part adopt 1,3,6,12 windows respectively
  • Figure 3: The framework of EGC
  • Figure 4: Distribution of METR-LA and PEMS-BAY monitoring sites
  • Figure 5: Forecasting Result from Graph-WaveNet and TAEGCN
  • ...and 2 more figures