DRFormer: Multi-Scale Transformer Utilizing Diverse Receptive Fields for Long Time-Series Forecasting

TL;DR

DRFormer is a multi-scale Transformer model with dynamic tokenization and position encoding, designed to improve long-term time series forecasting by capturing diverse temporal features across multiple resolutions.

Contribution

It introduces a dynamic sparse learning-based tokenizer, multi-scale Transformer architecture, and group-aware rotary position encoding for enhanced multi-resolution time series modeling.

Findings

Outperforms existing methods on real-world datasets

Effectively captures multi-scale temporal features

Demonstrates superior long-term forecasting accuracy

Abstract

Long-term time series forecasting (LTSF) has been widely applied in finance, traffic prediction, and other domains. Recently, patch-based transformers have emerged as a promising approach, segmenting data into sub-level patches that serve as input tokens. However, existing methods mostly rely on predetermined patch lengths, necessitating expert knowledge and posing challenges in capturing diverse characteristics across various scales. Moreover, time series data exhibit diverse variations and fluctuations across different temporal scales, which traditional approaches struggle to model effectively. In this paper, we propose a dynamic tokenizer with a dynamic sparse learning algorithm to capture diverse receptive fields and sparse patterns of time series data. In order to build hierarchical receptive fields, we develop a multi-scale Transformer model, coupled with multi-scale sequence extraction, capable of capturing multi-resolution features. Additionally, we introduce a group-aware rotary position encoding technique to enhance intra- and inter-group position awareness among representations across different temporal scales. Our proposed model, named DRFormer, is evaluated on various real-world datasets, and experimental results demonstrate its superiority compared to existing methods. Our code is available at: https://github.com/ruixindingECNU/DRFormer.