A lightweight Spatial-Temporal Graph Neural Network for Long-term Time Series Forecasting

TL;DR

Lite-STGNN is a lightweight, interpretable graph neural network that combines decomposition-based temporal modeling with learnable sparse spatial structures, achieving state-of-the-art long-term forecasting accuracy efficiently.

Contribution

The paper introduces Lite-STGNN, a novel lightweight spatial-temporal graph neural network that integrates decomposition and learnable sparse adjacency for improved long-term multivariate forecasting.

Findings

Achieves state-of-the-art accuracy on four benchmark datasets.

Parameter-efficient and faster training compared to transformer-based methods.

Spatial module improves forecast accuracy by 4.6%.

Abstract

We propose Lite-STGNN, a lightweight spatial-temporal graph neural network for long-term multivariate forecasting that integrates decomposition-based temporal modeling with learnable sparse graph structure. The temporal module applies trend-seasonal decomposition, while the spatial module performs message passing with low-rank Top-$K$ adjacency learning and conservative horizon-wise gating, enabling spatial corrections that enhance a strong linear baseline. Lite-STGNN achieves state-of-the-art accuracy on four benchmark datasets for horizons up to 720 steps, while being parameter-efficient and substantially faster to train than transformer-based methods. Ablation studies show that the spatial module yields 4.6% improvement over the temporal baseline, Top-$K$ enhances locality by 3.3%, and learned adjacency matrices reveal domain-specific interaction dynamics. Lite-STGNN thus offers a compact, interpretable, and efficient framework for long-term multivariate time series forecasting.