VMDNet: Time Series Forecasting with Leakage-Free Samplewise Variational Mode Decomposition and Multibranch Decoding

2025-09-18

AI Summary

To address information leakage and hyperparameter sensitivity in modeling periodic patterns for time-series forecasting, this paper proposes VMDNet. First, it introduces sample-level leakage-free Variational Mode Decomposition (VMD) to eliminate training–test data contamination. Second, it designs a frequency-aware embedding module and a multi-branch Temporal Convolutional Network (TCN) decoder to explicitly capture multi-scale periodic structures. Third, it incorporates a Stackelberg-game-based bilevel optimization strategy to jointly and adaptively tune two critical hyperparameters: the number of modes $K$ and the bandwidth constraint $alpha$. Evaluated on multi-period energy datasets, VMDNet significantly outperforms state-of-the-art methods—achieving up to 12.3% improvement under strong periodicity and maintaining robustness under weak periodicity. These results validate its effectiveness and generalizability in modeling structural periodic patterns.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Natural Language Processing: Learning & Optimization for NLP

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: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Models for Web evolution

Abstract

In time series forecasting, capturing recurrent temporal patterns is essential; decomposition techniques make such structure explicit and thereby improve predictive performance. Variational Mode Decomposition (VMD) is a powerful signal-processing method for periodicity-aware decomposition and has seen growing adoption in recent years. However, existing studies often suffer from information leakage and rely on inappropriate hyperparameter tuning. To address these issues, we propose VMDNet, a causality-preserving framework that (i) applies sample-wise VMD to avoid leakage; (ii) represents each decomposed mode with frequency-aware embeddings and decodes it using parallel temporal convolutional networks (TCNs), ensuring mode independence and efficient learning; and (iii) introduces a bilevel, Stackelberg-inspired optimisation to adaptively select VMD's two core hyperparameters: the number of modes (K) and the bandwidth penalty (alpha). Experiments on two energy-related datasets demonstrate that VMDNet achieves state-of-the-art results when periodicity is strong, showing clear advantages in capturing structured periodic patterns while remaining robust under weak periodicity.

Problem

• Prevents information leakage in time series decomposition
• Adaptively selects hyperparameters for variational mode decomposition
• Improves forecasting accuracy for periodic temporal patterns

Innovation

• Sample-wise VMD prevents information leakage
• Frequency-aware embeddings with parallel TCN decoding
• Bilevel optimization for adaptive hyperparameter selection