AWGformer: Adaptive Wavelet-Guided Transformer for Multi-Resolution Time Series Forecasting

TL;DR

AWGformer is a novel transformer-based architecture that combines adaptive wavelet decomposition and multi-scale attention mechanisms to improve multi-variate time series forecasting, especially for non-stationary signals.

Contribution

It introduces adaptive wavelet decomposition and frequency-aware attention within a transformer framework for enhanced multi-scale time series prediction.

Findings

Achieves significant improvements over state-of-the-art methods on benchmark datasets.

Effective in modeling multi-scale and non-stationary time series.

Provides theoretical guarantees and connects wavelet-guided attention to classical signal processing.

Abstract

Time series forecasting requires capturing patterns across multiple temporal scales while maintaining computational efficiency. This paper introduces AWGformer, a novel architecture that integrates adaptive wavelet decomposition with cross-scale attention mechanisms for enhanced multi-variate time series prediction. Our approach comprises: (1) an Adaptive Wavelet Decomposition Module (AWDM) that dynamically selects optimal wavelet bases and decomposition levels based on signal characteristics; (2) a Cross-Scale Feature Fusion (CSFF) mechanism that captures interactions between different frequency bands through learnable coupling matrices; (3) a Frequency-Aware Multi-Head Attention (FAMA) module that weights attention heads according to their frequency selectivity; (4) a Hierarchical Prediction Network (HPN) that generates forecasts at multiple resolutions before reconstruction. Extensive experiments on benchmark datasets demonstrate that AWGformer achieves significant average improvements over state-of-the-art methods, with particular effectiveness on multi-scale and non-stationary time series. Theoretical analysis provides convergence guarantees and establishes the connection between our wavelet-guided attention and classical signal processing principles.