Wavelet-based Disentangled Adaptive Normalization for Non-stationary Times Series Forecasting

2025-06-06

AI Summary

Nonstationary time series forecasting suffers from poor model generalizability due to time-varying statistical properties, yet existing methods often overlook their heterogeneous multi-component nonstationarity. To address this, we propose a model-agnostic, decoupled adaptive normalization framework. First, discrete wavelet transform decomposes the series into a trend component (strongly nonstationary) and a fluctuation component (weakly nonstationary). Second, distinct normalization strategies are applied: for the trend, first-order differencing stabilizes features to estimate dynamic normalization parameters; for the fluctuation, instance-level adaptive normalization is performed. This plug-and-play architecture requires no modification to backbone models. Empirically, it consistently improves forecasting accuracy across diverse state-of-the-art models—including Informer, Autoformer, and DLinear—on multiple benchmarks, demonstrating both broad applicability and effectiveness in mitigating nonstationarity-induced degradation.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior
• Search and Optimization: Non-convex Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Forecasting non-stationary time series is a challenging task because their statistical properties often change over time, making it hard for deep models to generalize well. Instance-level normalization techniques can help address shifts in temporal distribution. However, most existing methods overlook the multi-component nature of time series, where different components exhibit distinct non-stationary behaviors. In this paper, we propose Wavelet-based Disentangled Adaptive Normalization (WDAN), a model-agnostic framework designed to address non-stationarity in time series forecasting. WDAN uses discrete wavelet transforms to break down the input into low-frequency trends and high-frequency fluctuations. It then applies tailored normalization strategies to each part. For trend components that exhibit strong non-stationarity, we apply first-order differencing to extract stable features used for predicting normalization parameters. Extensive experiments on multiple benchmarks demonstrate that WDAN consistently improves forecasting accuracy across various backbone model. Code is available at this repository: https://github.com/MonBG/WDAN.

Problem

• Forecasting non-stationary time series with changing statistical properties
• Addressing multi-component non-stationary behaviors in time series
• Improving normalization for distinct trend and fluctuation components

Innovation

• Wavelet-based decomposition for time series components
• Tailored normalization for distinct non-stationary behaviors
• First-order differencing for stable trend feature extraction