PGN: The RNN's New Successor is Effective for Long-Range Time Series Forecasting

TL;DR

This work addresses the challenge of long-range time series forecasting with limited historical inputs by replacing recurrent propagation with a Parallel Gated Network (PGN) that uses a Historical Information Extraction layer and a single gate to achieve near $O(1)$ propagation across time steps, while maintaining $O(L)$ overall complexity. Building on PGN, Temporal PGN (TPGN) reshapes data into 2D and uses two branches to separately model long-term periodicity and short-term patterns, achieving $O(\sqrt{L})$ complexity for efficient forecasting. Empirical results on five real-world datasets show SOTA performance and strong efficiency, with ablations confirming the value of both branches and the PGN design. The work positions PGN as a practical successor to RNNs for long-range forecasting and lays a foundation for extending PGN to multivariate and broader temporal modeling tasks.

Abstract

Due to the recurrent structure of RNN, the long information propagation path poses limitations in capturing long-term dependencies, gradient explosion/vanishing issues, and inefficient sequential execution. Based on this, we propose a novel paradigm called Parallel Gated Network (PGN) as the new successor to RNN. PGN directly captures information from previous time steps through the designed Historical Information Extraction (HIE) layer and leverages gated mechanisms to select and fuse it with the current time step information. This reduces the information propagation path to $\mathcal{O}(1)$, effectively addressing the limitations of RNN. To enhance PGN's performance in long-range time series forecasting tasks, we propose a novel temporal modeling framework called Temporal PGN (TPGN). TPGN incorporates two branches to comprehensively capture the semantic information of time series. One branch utilizes PGN to capture long-term periodic patterns while preserving their local characteristics. The other branch employs patches to capture short-term information and aggregate the global representation of the series. TPGN achieves a theoretical complexity of $\mathcal{O}(\sqrt{L})$, ensuring efficiency in its operations. Experimental results on five benchmark datasets demonstrate the state-of-the-art (SOTA) performance and high efficiency of TPGN, further confirming the effectiveness of PGN as the new successor to RNN in long-range time series forecasting. The code is available in this repository: \url{https://github.com/Water2sea/TPGN}.

Figures (23)

• Figure 1: The information propagation illustration of different models.
• Figure 2: The structures of PGN and TPGN.
• Figure 3: Experimental results with different forecasting lengths on the ECL dataset.
• Figure 4: Time and memory overhead of different models.
• Figure 5: Experimental results with different forecasting lengths on the Traffic dataset.