VMDNet: Temporal Leakage-Free Variational Mode Decomposition for Electricity Demand Forecasting

Jin Zheng; John Cartlidge; Ran Tao; Weibin Feng

REVIEW 4 major objections 4 minor 35 references

VMDNet: Temporal Leakage-Free Variational Mode Decomposition for Electricity Demand Forecasting

T0 review · 4 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read VMDNet claims a leakage-free, sample-wise variational mode decomposition network that outperforms seven state-of-the-art baselines on strongly periodic electricity demand forecasting.

desk verdict A genuinely leakage-free VMD forecasting method with solid empirical support on one dataset, but an overstated metadata abstract and an unmeasured mode-stability assumption. read the letter →

arxiv 2509.15394 v3 pith:JAGL7XYI submitted 2025-09-18 cs.LG

Weibin Feng , Ran Tao , John Cartlidge , Jin Zheng This is my paper

classification cs.LG MSC 68T0762M1094A12

keywords variationalmodedecompositiontimeseriesforecastingelectricitydemandinformationleakagetemporalconvolutionalnetworkbileveloptimizationfrequency-awareembeddingperiodicity

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces VMDNet, a forecasting model that first decomposes each input window separately with Variational Mode Decomposition, so no future information can enter the decomposition; then encodes each band-limited mode with a frequency-aware embedding and decodes it with its own temporal convolutional network; and finally selects VMD's two key hyperparameters through a Stackelberg-inspired bilevel search. The central claim is that when periodicity is strong, this causal, mode-disentangled design beats seven recent baselines on all tested horizons on the Electricity Demand dataset, and remains competitive when periodicity is weak. If true, it shows that explicitly separating oscillatory components before learning is a practical path to both accuracy and interpretability in short-term demand forecasting.

What carries the argument

Variational Mode Decomposition (VMD) applied sample-wise: each input window x^(b) is decomposed into K band-limited modes u_k^(b) with center frequencies omega_k^(b) by solving a constrained variational problem via ADMM, using only past observations. The architecture then uses mode-specific token and frequency embeddings plus shared time and positional embeddings, K parallel causal TCN decoders with dilated convolutions, and a two-layer MLP fusion. A Stackelberg bilevel search treats K as leader and alpha as follower, using a Forecastability Information Criterion and a Mutual Information Criterion to pick the pair that suits the downstream predictor.

What would settle it

On the Electricity Demand dataset, compute the standard deviation and label-switch rate of each mode's center frequency across all sliding windows. If modes frequently swap order or frequencies drift by more than the bandwidth spacing, VMDNet's fixed mode-aligned decoders are misaligned, and a permutation-invariant variant should match or beat it.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a leakage-free forecasting pipeline can be built by applying K-mode VMD independently to each sliding window of length P, embedding each mode's center frequency along with its amplitude, and running one lightweight dilated TCN per mode before fusing branch outputs. Because each window's decomposition uses only data available at the forecast time, the approach is causal by construction. Experimentally, VMDNet achieves the best MSE and MAE across horizons 96, 192, and 336 on the strongly periodic Electricity Demand dataset, and the best average MAE on the weakly periodic Electricity Production dataset, with ablations showing that removing VMD, pa

Load-bearing premise

The load-bearing premise is that applying K-mode VMD separately to each 336-point window yields stable, consistently ordered band-limited modes whose center frequencies do not drift or swap across windows, so that the same mode-specific embeddings and TCN branches stay aligned.

Editorial extensions

If this is right

VMDNet's sample-wise VMD guarantees that no future information leaks into the decomposition, so the reported gains on the Electricity Demand dataset are not an artifact of look-ahead bias.
Strongly periodic forecasting can be done with small causal TCN branches rather than large transformer backbones, since VMD already isolates the periodic structure.
The bilevel selection of K and alpha provides a forecasting-aware alternative to reconstruction-error or entropy-based VMD tuning.
The method transfers to weakly periodic data without collapsing, though linear baselines may win on long horizons.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If mode center frequencies are stable across windows, the same architecture could be applied to multivariate and cross-domain forecasting by sharing mode branches across related series; the paper leaves this extension explicitly to future work.
A testable extension is to make the FIC/MIC selection differentiable or to warm-start VMD with center frequencies from the previous window, which could reduce the per-window computational cost while preserving leakage-freedom.
The strongest open risk is mode identity drift: since the paper itself notes that short-window VMD often yields weakly separated modes whose identities drift, VMDNet's fixed mode ordering could misalign across windows; measuring per-window frequency variance on the benchmark data would settle how often this happens.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The core idea is clean: decompose each input window separately with VMD so no future information enters the decomposition, then give each mode its own embedding and TCN branch. The leakage-free claim is defensible, and the architecture is a sensible response to the known mode-drift problem in short-window decomposition.

What is actually new: sample-wise VMD per sliding window is a real fix for the causality violation that plagues most decomposition-based forecasting. The parallel per-mode TCN branches are a natural fit for band-limited modes, and the ablation shows they matter—removing them costs 0.044 MAE. The bilevel Stackelberg-style search over K and alpha is a bit gimmicky as a framing, but it is a reasonable way to tune two coupled hyperparameters, and the comparison against PSO helps. Code is available, which is a plus.

Soft spots, in order of size. First, the metadata abstract claims three datasets and consistent superiority, while the paper itself uses two datasets and the weak-periodicity results are mixed—DLinear beats VMDNet on the long horizon. That overstatement needs to be fixed. Second, the mode-stability stress test is legitimate: short-window VMD can produce drifting mode identities, and the paper never measures how stable the center frequencies are across windows. The frequency embedding is designed to compensate, but the ablation shows removing it costs little, which makes me wonder whether the per-mode branches are actually robust to identity swaps. I would ask for a simple experiment—track center frequencies across consecutive windows or compute a mode-permutation rate. Third, the authors tune hyperparameters on validation and apply minor adjustments per horizon, which introduces a mild selection-bias risk, though the five-seed reporting reduces the alarm.

Overall, this is honest progress in a specialized subfield. It deserves a serious referee, but the authors should be pushed on the abstract and on measuring mode stability before publication. For someone doing decomposition-based forecasting, this is a useful paper to know.

Referee Report

4 major / 4 minor

Summary. VMDNet proposes a univariate time-series forecasting framework built on sample-wise Variational Mode Decomposition (VMD). Each lookback window is decomposed independently into K band-limited modes, so no future information enters the decomposition; modes are encoded with mode-specific token and frequency embeddings plus shared temporal/positional embeddings, decoded by parallel dilated TCNs, and fused by an MLP. A Stackelberg-inspired bilevel search over the VMD hyperparameters K and alpha is introduced, using a Forecastability Information Criterion (FIC) and a Mutual Information Criterion (MIC). Experiments are reported on two Polish electricity datasets (demand and production) at horizons 96, 192, and 336, with seven baselines and an ablation study. The central claims are that decomposition is leakage-free, that mode-specific parallel decoding improves forecasting, and that the bilevel search selects effective VMD parameters.

Significance. If the claims hold, VMDNet makes a useful contribution by addressing information leakage in decomposition-based forecasting and by giving each mode an explicit spectral identity. The paper ships code and data, reports results over five seeds, and provides ablations of the main components, which supports reproducibility. The leakage-free property is structurally guaranteed by Eqs. (4)-(7), and the idea of frequency-aware, mode-aligned decoding is well motivated. However, the central architectural assumption---that mode identities are stable enough across sliding windows to justify per-mode encoders and decoders---is not tested. The empirical evidence is also narrower than the abstract claims: only two datasets are used, and the weak-periodicity results do not show consistent superiority. The ablation confounds architectural capacity with the mode-separation mechanism. These issues need to be resolved before the central claims can be considered established.

major comments (4)

[Section 1; Eqs. (9)-(12); Table 2] The architecture assigns a dedicated TokenEmbed_k, FreqEmbed_k, and TCN_k to each VMD mode index. This is meaningful only if mode k consistently corresponds to the same spectral band across windows. The paper itself warns that short-window VMD 'often yields weakly separated modes whose identities drift across samples' (Section 1), but no experiment measures center-frequency stability, mode permutation rates, or alignment quality. If identities swap, branch k receives mixed spectral content and the claimed alignment benefit is not guaranteed. Please add a quantitative analysis of omega^(k,b) distributions across windows (e.g., pairwise permutation alignment or spectral centroid drift), and test a frequency-conditioned shared decoder as an alternative to per-mode branches. This is load-bearing because the ablation attributes a large gain to parallel decoding.
[Table 2; Section 3.5] The ablation 'w/o parallel decoding' replaces K TCN branches with a single TCN operating on the mean of mode embeddings. This changes both the architecture and the parameter count, so the observed MAE drop (+0.044) could be due to reduced capacity rather than to the value of per-mode decoding. Similarly, 'w/o VMD' uses a single TCN on raw input with no capacity-matched control. To support the claim that mode-specific parallel decoding is responsible for the gain, please include a control with matched parameter count (e.g., a single TCN with K-times width, or a shared TCN applied to concatenated modes without mixing). Without this, the central design conclusion is confounded.
[Abstract vs. Section 3.1 and Table 1] The abstract states that experiments are conducted on 'three widely used electricity demand datasets' and that VMDNet 'consistently outperforms state-of-the-art baselines.' The full text and Section 3.1 describe only two datasets, and Table 1 shows that on Electricity Production VMDNet is not consistently best: DLinear achieves a lower average MSE (0.596 vs. 0.665) and wins the MSE at horizons 192 and 336. Please correct the abstract and calibrate the claims to the actual evidence, e.g., 'achieves state-of-the-art when periodicity is strong and remains competitive under weak periodicity.'
[Section 2.5, Eqs. (13)-(16); Table 2] The paper motivates FIC and MIC as forecasting-aware criteria for selecting K and alpha, but it is not shown that minimizing these criteria improves downstream forecasting. The final selection in Section 2.5 uses validation-set performance after multiple restarts, not FIC/MIC alone. The 'w/o bilevel searching' ablation replaces the proposed scheme with PSO minimizing envelope entropy, but does not isolate the contribution of the FIC/MIC objectives. Please report the correlation between FIC/MIC values and validation/test forecast error, and compare the final selected (K, alpha) against random search or a simple grid using the same validation re-ranking. This is needed to support the claim that the Stackelberg-inspired search is the relevant mechanism, rather than standard validation-based tuning.

minor comments (4)

[Table 2] Column headers appear swapped: the reported values (0.156 MSE, 0.261 MAE for the full model) match Table 1's 'MSE' and 'MAE' order, not 'Test MAE' followed by 'Test MSE' as printed. Please correct.
[Eq. (13) and Section 3.3] Eq. (13) uses AR order r in the FIC penalty, while Section 3.3 reports 'AR order p=2'. Please harmonize the notation.
[Figure 1] The schematic is difficult to read in the typeset version; the mode-branch structure and the fusion step should be more clearly labeled.
[Eq. (1)] The variational objective is typeset with several garbled symbols; please ensure the Hilbert transform and demodulation terms are rendered cleanly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: VMDNet's claims are tested against external baselines and its leakage-free design follows directly from the stated input-window equations.

full rationale

The paper's central claims are empirical and evaluated against seven external baselines on two datasets. The leakage-free property is derived directly from the problem definition: Eq. (4) defines each input window as x^(b) := (x_{t_b-P+1:t_b}), and Eq. (7) states the VMD reconstruction constraint on that same window, so the decomposition depends only on observations available at time t_b. This is a direct consequence of the equations, not a circular redefinition. The proposed Stackelberg-inspired hyperparameter search (Section 2.5) is a heuristic for choosing K and alpha; the paper explicitly notes it has no formal convergence guarantee and that the final selection is made by evaluating candidate pairs on the validation set, which is standard model selection rather than a fitted quantity being presented as a prediction. The reported test MSE/MAE are benchmark comparisons against external models, so the empirical claim is not derived from the paper's own criteria. There are no load-bearing self-citations; the references to prior work (VMD, TCN, baseline methods) are external and do not supply an unverified premise. The skeptical concern about mode identity drift across short windows is a robustness/correctness issue, not a circularity, because the paper does not define its success metric in terms of mode stability. Overall, no circular step could be identified by the stated criteria.

Assumptions & free parameters 7 free parameters · 4 assumptions · 2 invented entities

The central method rests on four assumptions: the stability of sample-wise VMD modes across windows, the validity of the FIC/MIC proxies, the absence of preprocessing leakage, and the correctness of the VMD solver. K and alpha are fitted to validation data, and the AR order in FIC is chosen by hand. The FIC and MIC criteria are new constructs with no external evidence supporting their validity.

free parameters (7)

K (number of VMD modes) = 4 (Demand), 3 (Production)
Selected by FIC/MIC bilevel search plus validation-set selection (Section 2.5, 3.3). Central fitted hyperparameter.
alpha (VMD bandwidth penalty) = 5661 (Demand), 3802 (Production)
Co-selected with K; directly controls mode bandwidth and is fit to validation data.
AR order r in FIC = 2
Chosen when computing FIC (Section 3.3); affects FIC values and therefore the selected K.
d_model = 64 (Demand), 128 (Production)
Embedding dimension, tuned on validation.
TCN widths = [32,64,64] (Demand), [64,64] (Production)
Tuned on validation; sets per-mode branch capacity.
Bilevel restarts = 20
Number of independent restarts for the heuristic equilibrium; the best validation pair is chosen, adding a selection component.
Batch size and learning rate = batch 64/256; lr 1e-3
Standard training hyperparameters tuned on validation.

assumptions (4)

standard math VMD optimization (Eqs. 1-2) solved by ADMM returns band-limited modes that approximately reconstruct the signal.
The paper relies on VMD as a black-box decomposition; mode quality is assumed.
domain assumption Sample-wise VMD on each P-length window yields stable mode identities and center frequencies across windows.
The architecture assigns mode-specific embeddings and decoders; if VMD modes drift or swap across windows, these weights misalign. The paper notes this risk in Section 1 but does not quantify it.
ad hoc to paper FIC and MIC criteria are meaningful proxies for downstream forecasting performance.
Defined in Eqs. 13-14 without external validation; used to select K and alpha.
domain assumption No additional preprocessing (normalization, scaling) leaks future information.
Section 3.3 does not describe preprocessing; if normalization uses full-series statistics, the leakage-free claim would be violated.

invented entities (2)

Forecastability Information Criterion (FIC)
purpose: Selects the leader K in the bilevel hyperparameter search
New criterion proposed in Eq. 13; no independent benchmark for its validity.
Mutual Information Criterion (MIC)
purpose: Selects the follower alpha for a given K
New criterion proposed in Eq. 14; no external validation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of VMDNet: Temporal Leakage-Free Variational Mode Decomposition for Electricity Demand Forecasting." pith.science (2026). https://pith.science/paper/JAGL7XYI

@misc{pith2026250915394,
  author       = {Pith},
  title        = {Pith review of: VMDNet: Temporal Leakage-Free Variational Mode Decomposition for Electricity Demand Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JAGL7XYI}},
  note         = {Machine review of arXiv:2509.15394}
}

read the original abstract

Accurate electricity demand forecasting is challenging due to the strong multi-periodicity of real-world demand series, which makes effective modeling of recurrent temporal patterns crucial. 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 temporal 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 Stackelberg game inspired bilevel scheme to guide the selection of VMD's two key hyperparameters. Experiments on three widely used electricity demand datasets show that VMDNet consistently outperforms state-of-the-art baselines.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 2 linked inside Pith

[1]

Its primary objective is to infer future dynamics based on historical observations [1]

INTRODUCTION Time series forecasting is a fundamental task with widespread appli- cations in energy, transportation, and finance. Its primary objective is to infer future dynamics based on historical observations [1]. Given the inherent dynamics and temporal dependencies of real-world sys- tems, accurate modeling of periodic patterns and multi-scale struc...
[2]

Frequency-aware embeddings are used to encode the centre frequencies obtained from sample-wise VMD, enabling the model to learn periodic representations directly from sliding- window samples and thus achieve leakage-free forecasting
[3]

Parallel temporal convolutional networks (TCNs) are em- ployed to decode the representations of each mode indepen- dently, ensuring that every mode (corresponding to a distinct arXiv:2509.15394v1 [cs.LG] 18 Sep 2025 frequency) is modelled separately without mixing informa- tion across different frequency scales

arXiv 2025
[4]

A bilevel Stackelberg-inspired optimisation is introduced, where the number of modesKis treated as the leader and the bandwidth penaltyαas the follower, better reflecting the intrinsic structure of VMD and guiding adaptive parameter selection
[5]

METHODOLOGY 2.1. Variational Mode Decomposition VMD [9] decomposes a signalf(t)intoKband-limited IMFs uk(t), each centred at a frequencyω k, by solving a constrained variational optimisation problem: min {uk},{ωk} ( α KX k=1 ∂t δ(t) + j πt ∗u k(t) e−jωkt 2 L2 ) (1) subject to the reconstruction constraint KX k=1 uk(t) =f(t).(2) Here,α >0penalises the mode...
[6]

Dataset We evaluate the proposed model on two datasets with different pe- riodic characteristics

EXPERIMENTS 3.1. Dataset We evaluate the proposed model on two datasets with different pe- riodic characteristics. (i) Poland power system demand (Electricity Demand), contains hourly records of average power consumption in GW, exhibits strong periodicity. (ii) Poland electricity production by biomass (Electricity Production), reports hourly net generatio...

2021
[7]

CONCLUSION In this paper, we proposed VMDNet, a forecasting model tailored for strongly periodic time series. VMDNet performs sample-wise decomposition, explicitly disentangling oscillatory components into band-limited modes and thereby providing a leakage-free and in- terpretable representation of periodic structures. Extensive experi- ments on two real-...
[8]

JC is supported by UKRI EPSRC Grant No

ACKNOWLEDGMENTS WF’s PhD research is supported by UKRI EPSRC Doctoral Train- ing Partnership EP/W524414/1. JC is supported by UKRI EPSRC Grant No. EP/Y028392/1: AI for Collective Intelligence (AI4CI). The authors acknowledge the use of resources provided by the Isam- bard 3 Tier-2 HPC Facility. Isambard 3 is hosted by the University of Bristol and operate...

Show all 35 references

[9]

Founda- tion models for time series analysis: A tutorial and survey,

Yuxuan Liang, Haomin Wen, Yuqi Nie, Yushan Jiang, Ming Jin, Dongjin Song, Shirui Pan, and Qingsong Wen, “Founda- tion models for time series analysis: A tutorial and survey,” in ACM SIGKDD. 8 2024, pp. 6555–6565, ACM

2024
[10]

Informer: Be- yond efficient transformer for long sequence time-series fore- casting,

Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang, “Informer: Be- yond efficient transformer for long sequence time-series fore- casting,” in35th AAAI. 2021, vol. 12B, pp. 11106–11115, AAAI

2021
[11]

TimeMixer: Decomposable multiscale mixing for time series forecasting,

Shiyu Wang, Haixu Wu, Xiaoming Shi, Tengge Hu, Huakun Luo, Lintao Ma, James Y . Zhang, and Jun Zhou, “TimeMixer: Decomposable multiscale mixing for time series forecasting,” in12th ICLR. 2024, ICLR

2024
[12]

Research on in- formation leakage in time series prediction based on empirical mode decomposition,

Xinyi Yang, Jingyi Li, and Xuchu Jiang, “Research on in- formation leakage in time series prediction based on empirical mode decomposition,”Scientific Reports, vol. 14, 12 2024

2024
[13]

FEDformer: Frequency enhanced decomposed transformer for long-term series forecasting,

Tian Zhou, Ziqing Ma, Qingsong Wen, Xue Wang, Liang Sun, and Rong Jin, “FEDformer: Frequency enhanced decomposed transformer for long-term series forecasting,” in39th ICML. 2022, vol. 162, pp. 27268–27286, ML Research Press

2022
[14]

TimesNet: Temporal 2d-variation modeling for general time series analysis,

Haixu Wu, Tengge Hu, Yong Liu, Hang Zhou, Jianmin Wang, and Mingsheng Long, “TimesNet: Temporal 2d-variation modeling for general time series analysis,” in11th ICLR. 2023, ICLR

2023
[15]

WPMixer: Efficient Multi-Resolution Mixing for Long-Term Time Series Forecasting,

Md Mahmuddun Nabi Murad, Mehmet Aktukmak, and Yasin Yilmaz, “WPMixer: Efficient Multi-Resolution Mixing for Long-Term Time Series Forecasting,” inAAAI. 4 2025, vol. 39, pp. 19572–19580, AAAI

2025
[16]

Long- range transformers for dynamic spatiotemporal forecasting,

Jake Grigsby, Zhe Wang, Nam Nguyen, and Yanjun Qi, “Long- range transformers for dynamic spatiotemporal forecasting,” arXiv:2109.12218, 2021

2021 arXiv
[17]

Varia- tional mode decomposition,

Konstantin Dragomiretskiy and Dominique Zosso, “Varia- tional mode decomposition,”IEEE Transactions on Signal Processing, vol. 62, pp. 531–544, 2 2014

2014
[18]

A variational mode decompoisition approach for analysis and forecasting of economic and financial time se- ries,

Salim Lahmiri, “A variational mode decompoisition approach for analysis and forecasting of economic and financial time se- ries,”Expert Systems with Applications, vol. 55, pp. 268–273, 8 2016

2016
[19]

A novel hybrid model based on VMD-WT and PCA-BP-RBF neural network for short-term wind speed forecasting,

Yagang Zhang, Bing Chen, Guifang Pan, and Yuan Zhao, “A novel hybrid model based on VMD-WT and PCA-BP-RBF neural network for short-term wind speed forecasting,”En- ergy Conversion and Management, vol. 195, pp. 180–197, 9 2019

2019
[20]

A novel hybrid decomposition-ensemble model based on VMD and HGWO for container throughput forecasting,

Mingfei Niu, Yueyong Hu, Shaolong Sun, and Yu Liu, “A novel hybrid decomposition-ensemble model based on VMD and HGWO for container throughput forecasting,”Applied Mathematical Modelling, vol. 57, pp. 163–178, 5 2018

2018
[21]

A review and discussion of decomposition-based hybrid mod- els for wind energy forecasting applications,

Zheng Qian, Yan Pei, Hamidreza Zareipour, and Niya Chen, “A review and discussion of decomposition-based hybrid mod- els for wind energy forecasting applications,”Applied Energy, vol. 235, pp. 939–953, 2019

2019
[22]

Adaptive VMD based optimized deep learning mixed kernel ELM autoencoder for single and multistep wind power forecasting,

Vijaya Krishna Rayi, S. P. Mishra, Jyotirmayee Naik, and P. K. Dash, “Adaptive VMD based optimized deep learning mixed kernel ELM autoencoder for single and multistep wind power forecasting,”Energy, vol. 244, 4 2022

2022
[23]

Underwa- ter acoustic signal denoising model based on secondary vari- ational mode decomposition,

Hong Yang, Wen shuai Shi, and Guo hui Li, “Underwa- ter acoustic signal denoising model based on secondary vari- ational mode decomposition,”Defence Technology, vol. 28, pp. 87–110, 10 2023

2023
[24]

A multiscale and multivari- able differentiated learning for carbon price forecasting,

Linfei Chen and Xuefeng Zhao, “A multiscale and multivari- able differentiated learning for carbon price forecasting,”En- ergy Economics, vol. 131, pp. 107353, 2024

2024
[25]

A hybrid model for carbon price forecastingusing garch and long short-term memory network,

Yumeng Huang, Xingyu Dai, Qunwei Wang, and Dequn Zhou, “A hybrid model for carbon price forecastingusing garch and long short-term memory network,”Applied Energy, vol. 285, pp. 116485, 2021

2021
[26]

A genetic algorithm optimized hybrid model for agricultural price forecasting based on VMD and LSTM network,

Kapil Choudhary, Girish Kumar Jha, Ronit Jaiswal, and Ra- jeev Ranjan Kumar, “A genetic algorithm optimized hybrid model for agricultural price forecasting based on VMD and LSTM network,”Scientific Reports, vol. 15, no. 1, pp. 9932, Mar. 2025

2025
[27]

Ultra-short-term wind power prediction based on pvmd-esma-delm,

Guoqing An, Libo Chen, Jianxin Tan, Ziyao Jiang, Zheng Li, and Hexu Sun, “Ultra-short-term wind power prediction based on pvmd-esma-delm,”Energy Reports, vol. 8, pp. 8574–8588, 2022

2022
[28]

Heinrich V on Stackelberg,Market structure and equilibrium, Springer Berlin Heidelberg, 2011

2011
[29]

An overview of bilevel optimization,

Beno ˆıt Colson, Patrice Marcotte, and Gilles Savard, “An overview of bilevel optimization,”Annals of Operations Re- search, vol. 153, pp. 235–256, 9 2007

2007
[30]

iTransformer: Inverted transformers are effective for time series forecasting,

Yong Liu, Tengge Hu, Haoran Zhang, Haixu Wu, Shiyu Wang, Lintao Ma, and Mingsheng Long, “iTransformer: Inverted transformers are effective for time series forecasting,” in12th ICLR. 2024, ICLR

2024
[31]

Crossformer: Transformer utilizing cross-dimension dependency for multivariate time se- ries forecasting,

Yunhao Zhang and Junchi Yan, “Crossformer: Transformer utilizing cross-dimension dependency for multivariate time se- ries forecasting,” inICLR, 2023

2023
[32]

Are transformers effective for time series forecasting?,

Ailing Zeng, Muxi Chen, Lei Zhang, and Qiang Xu, “Are transformers effective for time series forecasting?,” in37th AAAI. 6 2023, vol. 37, pp. 11121–11128, AAAI Press

2023
[33]

A time series is worth 64 words: Long-term forecasting with transformers,

Yuqi Nie, Nam H. Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam, “A time series is worth 64 words: Long-term forecasting with transformers,” in11th ICLR. 2023, ICLR

2023
[34]

Autoformer: Decomposition transformers with auto- correlation for long-term series forecasting,

Haixu Wu, Jiehui Xu, Jianmin Wang, and Mingsheng Long, “Autoformer: Decomposition transformers with auto- correlation for long-term series forecasting,” inNeuralIPS. 2021, vol. 27, pp. 22419–22430, NeuralIPS foundation

2021
[35]

Short-term natural gas load forecasting based on EL-VMD- Transformer-ResLSTM,

M. Zhao, G. Guo, L. Fan, L. Han, Q. Yu, and Z. Wang, “Short-term natural gas load forecasting based on EL-VMD- Transformer-ResLSTM,”Scientific Reports, vol. 14, no. 1, Sept. 2024

2024

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

Its primary objective is to infer future dynamics based on historical observations [1]

INTRODUCTION Time series forecasting is a fundamental task with widespread appli- cations in energy, transportation, and finance. Its primary objective is to infer future dynamics based on historical observations [1]. Given the inherent dynamics and temporal dependencies of real-world sys- tems, accurate modeling of periodic patterns and multi-scale struc...

[2] [2]

Frequency-aware embeddings are used to encode the centre frequencies obtained from sample-wise VMD, enabling the model to learn periodic representations directly from sliding- window samples and thus achieve leakage-free forecasting

[3] [3]

Parallel temporal convolutional networks (TCNs) are em- ployed to decode the representations of each mode indepen- dently, ensuring that every mode (corresponding to a distinct arXiv:2509.15394v1 [cs.LG] 18 Sep 2025 frequency) is modelled separately without mixing informa- tion across different frequency scales

arXiv 2025

[4] [4]

A bilevel Stackelberg-inspired optimisation is introduced, where the number of modesKis treated as the leader and the bandwidth penaltyαas the follower, better reflecting the intrinsic structure of VMD and guiding adaptive parameter selection

[5] [5]

METHODOLOGY 2.1. Variational Mode Decomposition VMD [9] decomposes a signalf(t)intoKband-limited IMFs uk(t), each centred at a frequencyω k, by solving a constrained variational optimisation problem: min {uk},{ωk} ( α KX k=1 ∂t δ(t) + j πt ∗u k(t) e−jωkt 2 L2 ) (1) subject to the reconstruction constraint KX k=1 uk(t) =f(t).(2) Here,α >0penalises the mode...

[6] [6]

Dataset We evaluate the proposed model on two datasets with different pe- riodic characteristics

EXPERIMENTS 3.1. Dataset We evaluate the proposed model on two datasets with different pe- riodic characteristics. (i) Poland power system demand (Electricity Demand), contains hourly records of average power consumption in GW, exhibits strong periodicity. (ii) Poland electricity production by biomass (Electricity Production), reports hourly net generatio...

2021

[7] [7]

CONCLUSION In this paper, we proposed VMDNet, a forecasting model tailored for strongly periodic time series. VMDNet performs sample-wise decomposition, explicitly disentangling oscillatory components into band-limited modes and thereby providing a leakage-free and in- terpretable representation of periodic structures. Extensive experi- ments on two real-...

[8] [8]

JC is supported by UKRI EPSRC Grant No

ACKNOWLEDGMENTS WF’s PhD research is supported by UKRI EPSRC Doctoral Train- ing Partnership EP/W524414/1. JC is supported by UKRI EPSRC Grant No. EP/Y028392/1: AI for Collective Intelligence (AI4CI). The authors acknowledge the use of resources provided by the Isam- bard 3 Tier-2 HPC Facility. Isambard 3 is hosted by the University of Bristol and operate...