Lightweight Channel-wise Dynamic Fusion Model: Non-stationary Time Series Forecasting via Entropy Analysis

2025-03-04

AI Summary

To address global temporal dependency breakdown, channel-wise dynamic information loss, and over-smoothing in non-stationary time series forecasting, this paper proposes the lightweight Channel Dynamic Fusion Model (CDFM). Methodologically, CDFM introduces variance as an interpretable metric of non-stationarity—its first use in this context—and employs a dual-branch predictor to jointly model stationary and non-stationary components. A novel channel selector adaptively recovers critical dynamics by leveraging non-stationarity intensity, inter-channel similarity, and distribution consistency. Furthermore, variance-driven dynamic fusion weights balance normalized predictability with preservation of original dynamic characteristics. Evaluated on seven benchmark datasets, CDFM consistently outperforms state-of-the-art methods, achieving superior trade-offs among prediction accuracy, generalization capability, and computational efficiency. It effectively mitigates over-smoothing, temporal dependency fragmentation, and channel-agnostic modeling.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Temporal Planning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Non-stationarity is an intrinsic property of real-world time series and plays a crucial role in time series forecasting. Previous studies primarily adopt instance normalization to attenuate the non-stationarity of original series for better predictability. However, instance normalization that directly removes the inherent non-stationarity can lead to three issues: (1) disrupting global temporal dependencies, (2) ignoring channel-specific differences, and (3) producing over-smoothed predictions. To address these issues, we theoretically demonstrate that variance can be a valid and interpretable proxy for quantifying non-stationarity of time series. Based on the analysis, we propose a novel lightweight extit{C}hannel-wise extit{D}ynamic extit{F}usion extit{M}odel ( extit{CDFM}), which selectively and dynamically recovers intrinsic non-stationarity of the original series, while keeping the predictability of normalized series. First, we design a Dual-Predictor Module, which involves two branches: a Time Stationary Predictor for capturing stable patterns and a Time Non-stationary Predictor for modeling global dynamics patterns. Second, we propose a Fusion Weight Learner to dynamically characterize the intrinsic non-stationary information across different samples based on variance. Finally, we introduce a Channel Selector to selectively recover non-stationary information from specific channels by evaluating their non-stationarity, similarity, and distribution consistency, enabling the model to capture relevant dynamic features and avoid overfitting. Comprehensive experiments on seven time series datasets demonstrate the superiority and generalization capabilities of CDFM.

Problem

• Addresses non-stationarity in time series forecasting.
• Proposes dynamic fusion model to recover intrinsic non-stationarity.
• Improves prediction by capturing global and channel-specific dynamics.

Innovation

• Dual-Predictor Module captures stable and dynamic patterns
• Fusion Weight Learner dynamically characterizes non-stationarity
• Channel Selector recovers non-stationary information selectively