arxiv: 2604.28149 · v1 · submitted 2026-04-30 · 💻 cs.LG

Recognition: unknown

Explainable Load Forecasting with Covariate-Informed Time Series Foundation Models

Matthias Hertel , Alexandra Nikoltchovska , Sebastian P\"utz , Ralf Mikut , Benjamin Sch\"afer , Veit Hagenmeyer

Authors on Pith no claims yet

Pith reviewed 2026-05-07 07:41 UTC · model grok-4.3

classification 💻 cs.LG

keywords time series foundation modelsSHAP explanationsload forecastingzero-shot forecastingenergy systemscovariate maskingexplainable AIday-ahead forecasting

0 comments

The pith

Time series foundation models match specialized transformers in zero-shot load forecasting while providing SHAP explanations aligned with domain knowledge on weather and calendar effects.

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

The paper shows that general-purpose time series foundation models can forecast day-ahead electricity loads for a transmission system operator without any task-specific training and still reach accuracy levels comparable to a transformer trained on years of local data. It develops a SHAP explanation method that exploits the models' flexibility with input lengths and covariates through selective temporal and covariate masking, allowing scalable computation of feature contributions. The resulting explanations indicate that the models correctly draw on weather variables and calendar information in ways consistent with established energy domain knowledge. This combination of competitive zero-shot performance and built-in interpretability positions foundation models as practical, transparent options for operational forecasting in power systems where trust and reliability are essential.

Core claim

Covariate-informed time series foundation models, when explained via an efficient temporal and covariate masking strategy for SHAP, achieve zero-shot predictive performance on day-ahead TSO load forecasting that is competitive with a supervised transformer trained on multiple years of data, and the explanations confirm appropriate use of weather and calendar covariates in line with domain expertise.

What carries the argument

A temporal and covariate masking strategy for SHAP computation that leverages TSFMs' support for variable context lengths and additional input features to enable efficient, scalable explanations of forecasts by selectively withholding time steps or covariates.

Load-bearing premise

The temporal and covariate masking strategy yields faithful and unbiased SHAP explanations of the TSFM predictions without introducing artifacts from the masking process or approximation errors.

What would settle it

A case where the SHAP values from the masking method assign high importance to irrelevant features or contradict known physical relationships, such as temperature effects on load, or where the explained predictions deviate substantially from the model's actual outputs when inputs are masked.

Figures

Figures reproduced from arXiv: 2604.28149 by Alexandra Nikoltchovska, Benjamin Sch\"afer, Matthias Hertel, Ralf Mikut, Sebastian P\"utz, Veit Hagenmeyer.

**Figure 1.** Figure 1: Overview of our approach to efficiently compute exact Shapley Additive Explanations (SHAP) for covariate-informed view at source ↗

**Figure 2.** Figure 2: Dependence of Chronos-2 (left) and TabPFN-TS (right) on the holiday covariate (top), the temperature covariate view at source ↗

**Figure 3.** Figure 3: Local SHAP explanations for Chronos-2 predictions on selected days showing feature contributions to load forecasts: (a) view at source ↗

**Figure 4.** Figure 4: Performance comparison of the TabPFN-TS and view at source ↗

**Figure 5.** Figure 5: SHAP dependence plots of temperature difference between prediction day and previous day. The back line indicates view at source ↗

**Figure 6.** Figure 6: Local SHAP explanations for TabPFN-TS predictions on selected days showing feature contributions to load forecasts: view at source ↗

**Figure 7.** Figure 7: Chronos-2 predictions and local SHAP feature attributions across the full test set, with one panel per month. Each view at source ↗

**Figure 8.** Figure 8: TabPFN-TS predictions and local SHAP feature attributions across the full test set, with one panel per month. Each view at source ↗

read the original abstract

Time Series Foundation Models (TSFMs) have recently emerged as general-purpose forecasting models and show considerable potential for applications in energy systems. However, applications in critical infrastructure like power grids require transparency to ensure trust and reliability and cannot rely on pure black-box models. To enhance the transparency of TSFMs, we propose an efficient algorithm for computing Shapley Additive Explanations (SHAP) tailored to these models. The proposed approach leverages the flexibility of TSFMs with respect to input context length and provided covariates. This property enables efficient temporal and covariate masking (selectively withholding inputs), allowing for a scalable explanation of model predictions using SHAP. We evaluate two TSFMs - Chronos-2 and TabPFN-TS - on a day-ahead load forecasting task for a transmission system operator (TSO). In a zero-shot setting, both models achieve predictive performance competitive with a Transformer model trained specifically on multiple years of TSO data. The explanations obtained through our proposed approach align with established domain knowledge, particularly as the TSFMs appropriately use weather and calendar information for load prediction. Overall, we demonstrate that TSFMs can serve as transparent and reliable tools for operational energy forecasting.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper adapts SHAP to TSFMs via temporal and covariate masking for zero-shot load forecasting, gets explanations that match domain knowledge on weather and calendars, but the faithfulness of those attributions under context changes is not strongly validated.

read the letter

The key takeaway is that this paper shows how to make time series foundation models more explainable for energy load forecasting by using an efficient SHAP method based on temporal and covariate masking. Two models, Chronos-2 and TabPFN-TS, achieve zero-shot performance competitive with a dedicated Transformer on TSO data, and their explanations correctly highlight the role of weather and calendar information. The strength of the work is in the practical application. Power grid operators need reliable forecasts but also need to understand why a model predicts a certain load. By leveraging the models' flexibility with input lengths, the masking allows scalable computation of attributions without retraining. This is a direct response to the need for transparency in critical infrastructure, and the alignment with domain knowledge is a positive sign that the models are picking up real patterns rather than artifacts. What the paper does well is demonstrate that general-purpose TSFMs can be adapted for operational use with added explainability. The zero-shot aspect is particularly appealing because it reduces the need for large local datasets. The soft spots are around verification. The abstract gives no numbers, so we cannot see the actual error rates or how close the performance really is. The central concern is whether the SHAP values are faithful. Masking inputs changes the context the model sees, which for autoregressive models might not satisfy the assumptions behind SHAP additivity. Without tests like faithfulness metrics or synthetic experiments to confirm the attributions recover known effects, the alignment with domain knowledge could be partly due to the method rather than true model behavior. If the full paper includes such checks, that would address it; otherwise it remains a gap. This paper is for researchers and practitioners in energy forecasting who want to use foundation models but require interpretability. A reader working on XAI for time series would get concrete ideas from the masking strategy. It deserves serious peer review. The idea is timely and the execution seems competent, even if more evidence on explanation quality is needed. I recommend sending it to referees with a note to strengthen the validation of the SHAP procedure.

Referee Report

3 major / 3 minor

Summary. The paper introduces an efficient algorithm for computing SHAP explanations for Time Series Foundation Models (TSFMs) by using temporal and covariate masking to handle variable context lengths. It evaluates two TSFMs, Chronos-2 and TabPFN-TS, on a zero-shot day-ahead load forecasting task using data from a transmission system operator (TSO). The models are shown to achieve predictive performance competitive with a Transformer trained on multiple years of data, and the explanations are claimed to align with domain knowledge, particularly in the use of weather and calendar information.

Significance. If the central claims hold, this work is significant because it addresses the transparency requirement for using foundation models in critical infrastructure applications like energy load forecasting. The proposed masking-based SHAP method exploits the unique properties of TSFMs to provide scalable explanations without the need for model-specific adaptations. This could encourage the use of pretrained models in operational settings where both accuracy and interpretability are essential. The zero-shot capability is a notable strength if supported by robust comparisons.

major comments (3)

[Section 3.2] Section 3.2: The temporal and covariate masking strategy for SHAP computation assumes that the TSFM's predictive function remains additive and consistent under masking. However, for models like Chronos-2 which are autoregressive and pretrained on fixed or variable contexts, selectively withholding past timesteps or covariates changes the effective input distribution, which may cause the model to impute or re-encode the context in ways that violate SHAP's assumptions. The manuscript does not provide quantitative fidelity checks, such as recovery of known feature importances on synthetic data or perturbation-based faithfulness metrics, to rule out systematic artifacts in the reported explanations.
[Section 4.1] Section 4.1: The performance comparison in the zero-shot setting is described as 'competitive' but lacks specific quantitative metrics (e.g., MAE, RMSE values), error bars, or statistical significance tests against the supervised Transformer baseline. Without these, it is difficult to verify the claim that the TSFMs perform on par with a model trained on multiple years of TSO data, which is central to arguing for their reliability.
[Section 4.3] Section 4.3: The alignment of explanations with domain knowledge is presented qualitatively. To strengthen the claim that TSFMs are 'transparent and reliable,' the paper should include an ablation or sensitivity analysis showing that the SHAP values are stable and not sensitive to the choice of baseline or masking order.

minor comments (3)

[Figure 2] The explanation plots would benefit from clearer legends indicating the meaning of positive and negative SHAP values in the context of load forecasting.
[Notation] Define the notation for the masked inputs more explicitly in the method section to avoid ambiguity in how the baseline is chosen for SHAP.
[References] Consider adding references to recent work on explainability for time series models and foundation models to better position the contribution.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments on our work. These suggestions will help us clarify the methodological assumptions, strengthen the empirical comparisons, and provide additional robustness checks for the explanations. Below, we respond to each major comment and indicate the revisions we plan to incorporate.

read point-by-point responses

Referee: [Section 3.2] Section 3.2: The temporal and covariate masking strategy for SHAP computation assumes that the TSFM's predictive function remains additive and consistent under masking. However, for models like Chronos-2 which are autoregressive and pretrained on fixed or variable contexts, selectively withholding past timesteps or covariates changes the effective input distribution, which may cause the model to impute or re-encode the context in ways that violate SHAP's assumptions. The manuscript does not provide quantitative fidelity checks, such as recovery of known feature importances on synthetic data or perturbation-based faithfulness metrics, to rule out systematic artifacts in the reported explanations.

Authors: We thank the referee for highlighting this important consideration regarding the application of SHAP to autoregressive TSFMs. Our masking strategy is intended to exploit the models' flexibility with variable context lengths, but we recognize that autoregressive decoding may introduce non-additive effects. To rigorously address this, we will include quantitative fidelity checks in the revised manuscript. Specifically, we will conduct experiments on synthetic datasets with known ground-truth importances and report perturbation-based faithfulness metrics to validate that the explanations do not contain systematic artifacts. revision: yes
Referee: [Section 4.1] Section 4.1: The performance comparison in the zero-shot setting is described as 'competitive' but lacks specific quantitative metrics (e.g., MAE, RMSE values), error bars, or statistical significance tests against the supervised Transformer baseline. Without these, it is difficult to verify the claim that the TSFMs perform on par with a model trained on multiple years of TSO data, which is central to arguing for their reliability.

Authors: We agree that more detailed quantitative metrics are necessary to support the competitiveness claim. Although the manuscript describes the performance as competitive, we will revise Section 4.1 to include specific MAE and RMSE values for all models, along with error bars derived from multiple evaluation runs and results from statistical significance tests (such as Wilcoxon signed-rank tests) comparing the TSFMs to the supervised Transformer baseline. revision: yes
Referee: [Section 4.3] Section 4.3: The alignment of explanations with domain knowledge is presented qualitatively. To strengthen the claim that TSFMs are 'transparent and reliable,' the paper should include an ablation or sensitivity analysis showing that the SHAP values are stable and not sensitive to the choice of baseline or masking order.

Authors: We acknowledge that the current analysis of explanation alignment is primarily qualitative. To provide stronger evidence for the reliability of the explanations, we will add an ablation and sensitivity analysis in the revised Section 4.3. This will include testing the stability of SHAP values across different baseline choices and variations in masking order, with quantitative metrics to demonstrate robustness. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on independent empirical evaluation and external domain knowledge

full rationale

The paper proposes a masking-based SHAP procedure for TSFMs, then reports zero-shot forecasting performance that is compared to an independently trained Transformer baseline on TSO data and validates that the resulting attributions align with established external domain knowledge on weather/calendar effects. No equations or steps reduce the central claims (performance competitiveness or explanation faithfulness) to self-definition, fitted inputs renamed as predictions, or load-bearing self-citations. The derivation chain consists of a new algorithmic adaptation followed by standard empirical benchmarking; it does not contain self-referential reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work relies on standard assumptions from explainable AI and time series modeling without introducing new free parameters or invented entities.

axioms (2)

standard math SHAP values provide additive feature attributions that sum to the model output difference
Core mathematical property of SHAP invoked for the explanation method.
domain assumption TSFMs flexibly accept varying context lengths and covariate inputs
This property is used to enable the temporal and covariate masking strategy.

pith-pipeline@v0.9.0 · 5529 in / 1389 out tokens · 77792 ms · 2026-05-07T07:41:29.531869+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

53 extracted references · 43 canonical work pages · 3 internal anchors

[1]

Taha Aksu, Gerald Woo, Juncheng Liu, Xu Liu, Chenghao Liu, Silvio Savarese, Caiming Xiong, and Doyen Sahoo. 2024. GIFT-Eval: A Benchmark For Gen- eral Time Series Forecasting Model Evaluation. doi:10.48550/arXiv.2410.10393 arXiv:2410.10393 [cs]

work page doi:10.48550/arxiv.2410.10393 2024
[2]

Chronos-2: From Univariate to Universal Forecasting

Abdul Fatir Ansari, Oleksandr Shchur, Jaris Küken, Andreas Auer, Boran Han, Pedro Mercado, Syama Sundar Rangapuram, Huibin Shen, Lorenzo Stella, Xiyuan Zhang, Mononito Goswami, Shubham Kapoor, Danielle C. Maddix, M. Hertel, A. Nikoltchovska, et al. Pablo Guerron, Tony Hu, Junming Yin, Nick Erickson, Prateek Mutalik De- sai, Hao Wang, Huzefa Rangwala, Geor...

work page internal anchor Pith review doi:10.48550/arxiv.2510.15821 2025
[3]

Chronos: Learning the Language of Time Series

Abdul Fatir Ansari, Lorenzo Stella, Caner Turkmen, Xiyuan Zhang, Pedro Mercado, Huibin Shen, Oleksandr Shchur, Syama Sundar Rangapuram, Sebas- tian Pineda Arango, Shubham Kapoor, Jasper Zschiegner, Danielle C. Maddix, Hao Wang, Michael W. Mahoney, Kari Torkkola, Andrew Gordon Wilson, Michael Bohlke-Schneider, and Yuyang Wang. 2024. Chronos: Learning the L...

work page internal anchor Pith review doi:10.48550/arxiv.2403.07815 2024
[4]

Lukas Baur, Konstantin Ditschuneit, Maximilian Schambach, Can Kaymakci, Thomas Wollmann, and Alexander Sauer. 2024. Explainability and Interpretability in Electric Load Forecasting Using Machine Learning Techniques – A Review. Energy and AI16 (May 2024), 100358. doi:10.1016/j.egyai.2024.100358

work page doi:10.1016/j.egyai.2024.100358 2024
[5]

Cruz, Mário A.T

João Bento, Pedro Saleiro, André F. Cruz, Mário A.T. Figueiredo, and Pedro Bizarro. 2021. TimeSHAP: Explaining Recurrent Models through Sequence Perturbations. InProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining (KDD ’21). Association for Computing Machinery, New York, NY, USA, 2565–2573. doi:10.1145/3447548.3467166

work page doi:10.1145/3447548.3467166 2021
[6]

Adrien Bibal, Rémi Cardon, David Alfter, Rodrigo Wilkens, Xiaoou Wang, Thomas François, and Patrick Watrin. 2022. Is Attention Explanation? An Intro- duction to the Debate. InProceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Smaranda Muresan, Preslav Nakov, and Aline Villavicencio (Eds.). Asso...

work page doi:10.18653/v1/2022.acl-long.269 2022
[7]

Matthieu Boileau, Philippe Helluy, Jeremy Pawlus, and Svitlana Vyetrenko. 2025. Towards Interpretable Time Series Foundation Models. doi:10.48550/arXiv.2507. 07439 arXiv:2507.07439 [cs]

work page doi:10.48550/arxiv.2507 2025
[8]

G. E. P. Box and D. R. Cox. 1964. An Analysis of Transformations.Journal of the Royal Statistical Society: Series B (Methodological)26, 2 (July 1964), 211–243. doi:10.1111/j.2517-6161.1964.tb00553.x

work page doi:10.1111/j.2517-6161.1964.tb00553.x 1964
[9]

Ben Cohen, Emaad Khwaja, Youssef Doubli, Salahidine Lemaachi, Chris Lettieri, Charles Masson, Hugo Miccinilli, Elise Ramé, Qiqi Ren, Afshin Rostamizadeh, Jean Ogier du Terrail, Anna-Monica Toon, Kan Wang, Stephan Xie, Zongzhe Xu, Viktoriya Zhukova, David Asker, Ameet Talwalkar, and Othmane Abou- Amal. 2025. This Time is Different: An Observability Perspec...

work page doi:10.48550/arxiv.2505.14766 2025
[10]

Abhimanyu Das, Weihao Kong, Rajat Sen, and Yichen Zhou. 2024. A decoder-only foundation model for time-series forecasting. http://arxiv.org/abs/2310.10688 arXiv:2310.10688 [cs]

work page arXiv 2024
[11]

Björn Deiseroth, Mayukh Deb, Samuel Weinbach, Manuel Brack, Patrick Schramowski, and Kristian Kersting. 2023. ATMAN: Understanding Transformer Predictions Through Memory Efficient Attention Manipula- tion.Advances in Neural Information Processing Systems36 (Dec. 2023), 63437–63460. https://proceedings.neurips.cc/paper_files/paper/2023/hash/ c83bc020a020cd...

2023
[12]

ENTSO-E. 2025. Transparency platform. https://transparency.entsoe.eu

2025
[13]

European Commission. 2024. Artificial Intelligence Act. https://eur-lex.europa. eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202401689#art_13

2024
[14]

Elena Giacomazzi, Felix Haag, and Konstantin Hopf. 2023. Short-Term Elec- tricity Load Forecasting Using the Temporal Fusion Transformer: Effect of Grid Hierarchies and Data Sources. InProceedings of the 14th ACM Interna- tional Conference on Future Energy Systems. ACM, Orlando FL USA, 353–360. doi:10.1145/3575813.3597345

work page doi:10.1145/3575813.3597345 2023
[15]

Miha Grabner, Yi Wang, Qingsong Wen, Boštjan Blažič, and Vitomir Štruc. 2023. A Global Modeling Framework for Load Forecasting in Distribution Networks. IEEE Transactions on Smart Grid14, 6 (Nov. 2023), 4927–4941. doi:10.1109/TSG. 2023.3264525

work page doi:10.1109/tsg 2023
[16]

Stephen Haben, Siddharth Arora, Georgios Giasemidis, Marcus Voss, and Danica Vukadinović Greetham. 2021. Review of low voltage load forecasting: Methods, applications, and recommendations.Applied Energy304 (Dec. 2021), 117798. doi:10.1016/j.apenergy.2021.117798

work page doi:10.1016/j.apenergy.2021.117798 2021
[17]

Hans Hersbach, Bill Bell, Paul Berrisford, Shoji Hirahara, András Horányi, Joaquín Muñoz-Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Dinand Schepers, Adrian Simmons, Cornel Soci, Saleh Abdalla, Xavier Abellan, Gian- paolo Balsamo, Peter Bechtold, Gionata Biavati, Jean Bidlot, Massimo Bonavita, Giovanna De Chiara, Per Dahlgren, Dick Dee, Michail D...

work page doi:10.1002/qj.3803 2020
[18]

Matthias Hertel, Lara Ambrosius, Manuel Treutlein, Ralf Mikut, and Veit Ha- genmeyer. 2025. A comparison of local, cluster-specific and global Transformer models for forecasting electrical loads of individual buildings and substations. In 2025 IEEE Kiel PowerTech. IEEE, Kiel, Germany, 1–8. doi:10.1109/PowerTech59965. 2025.11180482

work page doi:10.1109/powertech59965 2025
[19]

Matthias Hertel, Maximilian Beichter, Benedikt Heidrich, Oliver Neumann, Ben- jamin Schäfer, Ralf Mikut, and Veit Hagenmeyer. 2023. Transformer training strategies for forecasting multiple load time series.Energy Informatics6, 1 (Oct. 2023), 20. doi:10.1186/s42162-023-00278-z

work page doi:10.1186/s42162-023-00278-z 2023
[20]

Matthias Hertel, Simon Ott, Benjamin Schäfer, Ralf Mikut, Veit Hagenmeyer, and Oliver Neumann. 2022. Evaluation of Transformer Architectures for Electrical Load Time-Series Forecasting. InProceedings 32. Workshop Computational Intelli- gence, Vol. 1. 93. https://library.oapen.org/bitstream/handle/20.500.12657/59840/ external_content.pdf?sequence=1#page=105

2022
[21]

Matthias Hertel, Sebastian Pütz, Ralf Mikut, Veit Hagenmeyer, and Benjamin Schäfer. 2025. Explainable time-series forecasting with sampling-free SHAP for Transformers. doi:10.48550/arXiv.2512.20514 arXiv:2512.20514 [cs]

work page doi:10.48550/arxiv.2512.20514 2025
[22]

Noah Hollmann, Samuel Müller, Katharina Eggensperger, and Frank Hutter. 2023. TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second. doi:10.48550/arXiv.2207.01848 arXiv:2207.01848 [cs]

work page internal anchor Pith review doi:10.48550/arxiv.2207.01848 2023
[23]

Tao Hong. 2014. Energy Forecasting: Past, Present, and Future.Foresight: The International Journal of Applied Forecasting32 (2014), 43–48. https://ideas.repec. org//a/for/ijafaa/y2014i32p43-48.html

2014
[24]

Tao Hong, Pierre Pinson, Yi Wang, Rafał Weron, Dazhi Yang, and Hamidreza Zareipour. 2020. Energy Forecasting: A Review and Outlook.IEEE Open Access Journal of Power and Energy7 (2020), 376–388. doi:10.1109/OAJPE.2020.3029979

work page doi:10.1109/oajpe.2020.3029979 2020
[25]

Shi Bin Hoo, Samuel Müller, David Salinas, and Frank Hutter. 2025. From Tables to Time: How TabPFN-v2 Outperforms Specialized Time Series Forecasting Models. doi:10.48550/arXiv.2501.02945 arXiv:2501.02945 [cs]

work page doi:10.48550/arxiv.2501.02945 2025
[26]

Sarthak Jain and Byron C. Wallace. 2019. Attention is not Explanation. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), Jill Burstein, Christy Doran, and Thamar Solorio (Eds.). Association for Computational Linguistics, Minne...

2019
[27]

Nouha Karaouli, Denis Coquenet, Elisa Fromont, Martial Mermillod, and Ma- rina Reyboz. 2025. How Foundational are Foundation Models for Time Series Forecasting? doi:10.48550/arXiv.2510.00742 arXiv:2510.00742 [cs]

work page doi:10.48550/arxiv.2510.00742 2025
[28]

Khan, and Naveed Ahmed Khan

Ahsan Raza Khan, Anzar Mahmood, Awais Safdar, Zafar A. Khan, and Naveed Ahmed Khan. 2016. Load forecasting, dynamic pricing and DSM in smart grid: A review.Renewable and Sustainable Energy Reviews54 (Feb. 2016), 1311–1322. doi:10.1016/j.rser.2015.10.117

work page doi:10.1016/j.rser.2015.10.117 2016
[29]

2014.Load management as a way of covering peak demand in Southern Germany

Marian Klobasa, Gerhard Angerer, Arne Lüllmann, Joachim Schleich, Tim Buber, Anna Gruber, Marie Hünecke, and Serafin von Roon. 2014.Load management as a way of covering peak demand in Southern Germany. Final Report 040/04-S- 2014/EN. Fraunhofer Institute for Systems and Innovation Research (ISI) and Forschungsgesellschaft für Energiewirtschaft mbH. doi:10...

work page doi:10.24406/publica-fhg- 2014
[30]

Siva Rama Krishna Kottapalli, Karthik Hubli, Sandeep Chandrashekhara, Garima Jain, Sunayana Hubli, Gayathri Botla, and Ramesh Doddaiah. 2025. Foun- dation Models for Time Series: A Survey. doi:10.48550/arXiv.2504.04011 arXiv:2504.04011 [cs]

work page doi:10.48550/arxiv.2504.04011 2025
[31]

Alexander Kreusel, Matthias Hertel, Moritz Noskiewicz, Heiko Maaß, Ralf Mikut, and Veit Hagenmeyer. 2025. Evaluating Time-Series Foundation Models for Cooling Demand Forecasting with Little Data. InProceedings 35. Workshop Com- putational Intelligence. Berlin, 301–322

2025
[32]

Yuxuan Liang, Haomin Wen, Yuqi Nie, Yushan Jiang, Ming Jin, Dongjin Song, Shirui Pan, and Qingsong Wen. 2024. Foundation Models for Time Series Analysis: A Tutorial and Survey. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 6555–6565. doi:10.1145/3637528.3671451 arXiv:2403.14735 [cs]

work page doi:10.1145/3637528.3671451 2024
[33]

Nan Lin, Dong Yun, Weijie Xia, Peter Palensky, and Pedro P. Vergara. 2025. Comparative Analysis of Zero-Shot Capability of Time-Series Foundation Models in Short-Term Load Prediction. In2025 IEEE Kiel PowerTech. IEEE, Kiel, Germany, 1–6. doi:10.1109/PowerTech59965.2025.11180251

work page doi:10.1109/powertech59965.2025.11180251 2025
[34]

Chenghao Liu, Taha Aksu, Juncheng Liu, Xu Liu, Hanshu Yan, Quang Pham, Silvio Savarese, Doyen Sahoo, Caiming Xiong, and Junnan Li. 2026. Moirai 2.0: When Less Is More for Time Series Forecasting. doi:10.48550/arXiv.2511.11698 arXiv:2511.11698 [cs]

work page doi:10.48550/arxiv.2511.11698 2026
[35]

Scott M Lundberg and Su-In Lee. 2017. A Unified Approach to Interpreting Model Predictions. InAdvances in Neural Information Processing Systems, Vol. 30. Canada, 4768 – 4777. https://dl.acm.org/doi/proceedings/10.5555/3295222

work page doi:10.5555/3295222 2017
[36]

Machlev, L

R. Machlev, L. Heistrene, M. Perl, K. Y. Levy, J. Belikov, S. Mannor, and Y. Levron
[37]

2022), 100169

Explainable Artificial Intelligence (XAI) techniques for energy and power systems: Review, challenges and opportunities.Energy and AI9 (Aug. 2022), 100169. doi:10.1016/j.egyai.2022.100169

work page doi:10.1016/j.egyai.2022.100169 2022
[38]

Marcel Meyer, Sascha Kaltenpoth, Kevin Zalipski, and Oliver Müller. 2025. Time Series Foundation Models: Benchmarking Challenges and Requirements. doi:10. 48550/arXiv.2510.13654 arXiv:2510.13654 [cs]

work page arXiv 2025
[39]

Marcel Meyer, David Zapata, Sascha Kaltenpoth, and Oliver Müller. 2024. Bench- marking Time Series Foundation Models for Short-Term Household Electricity Load Forecasting. doi:10.48550/arXiv.2410.09487 arXiv:2410.09487. Explainable Load Forecasting with Covariate-Informed Time Series Foundation Models

work page doi:10.48550/arxiv.2410.09487 2024
[40]

Pablo Montero-Manso and Rob J. Hyndman. 2021. Principles and algorithms for forecasting groups of time series: Locality and globality.International Journal of Forecasting37, 4 (Oct. 2021), 1632–1653. doi:10.1016/j.ijforecast.2021.03.004

work page doi:10.1016/j.ijforecast.2021.03.004 2021
[41]

Reddy, Brandon Foreman, and Vignesh Subbian

Amin Nayebi, Sindhu Tipirneni, Chandan K. Reddy, Brandon Foreman, and Vignesh Subbian. 2023. WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values.Journal of Biomedical Informatics 144 (Aug. 2023), 104438. doi:10.1016/j.jbi.2023.104438

work page doi:10.1016/j.jbi.2023.104438 2023
[42]

Alexandra Nikoltchovska, Sebastian Pütz, Xiao Li, Veit Hagenmeyer, and Ben- jamin Schäfer. 2025. Probabilistic and Explainable Machine Learning for Tabular Power Grid Data. InProceedings of the 16th ACM International Conference on Future and Sustainable Energy Systems (E-Energy ’25). Association for Computing Machinery, New York, NY, USA, 213–231. doi:10....

work page doi:10.1145/3679240.3734623 2025
[43]

Atharva Pandey, Abhilash Neog, and Gautam Jajoo. 2025. On the Internal Semantics of Time-Series Foundation Models. doi:10.48550/ARXIV.2511.15324 Version Number: 1

work page doi:10.48550/arxiv.2511.15324 2025
[44]

Kashif Rasul, Arjun Ashok, Andrew Robert Williams, Hena Ghonia, Rishika Bhag- watkar, Arian Khorasani, Mohammad Javad Darvishi Bayazi, George Adamopou- los, Roland Riachi, Nadhir Hassen, Marin Biloš, Sahil Garg, Anderson Schneider, Nicolas Chapados, Alexandre Drouin, Valentina Zantedeschi, Yuriy Nevmyvaka, and Irina Rish. 2023. Lag-Llama: Towards Foundati...

work page doi:10.48550/arxiv.2310.08278 2023
[45]

David Rundel, Julius Kobialka, Constantin von Crailsheim, Matthias Feurer, Thomas Nagler, and David Rügamer. 2024. Interpretable Machine Learning for TabPFN. InExplainable Artificial Intelligence, Luca Longo, Sebastian La- puschkin, and Christin Seifert (Eds.). Vol. 2154. Springer Nature Switzerland, Cham, 465–476. doi:10.1007/978-3-031-63797-1_23 Series ...

work page doi:10.1007/978-3-031-63797-1_23 2024
[46]

Frederik vom Scheidt, Hana Medinová, Nicole Ludwig, Bent Richter, Philipp Staudt, and Christof Weinhardt. 2020. Data analytics in the electricity sector – A quantitative and qualitative literature review.Energy and AI1 (Aug. 2020), 100009. doi:10.1016/j.egyai.2020.100009

work page doi:10.1016/j.egyai.2020.100009 2020
[47]

Shapley, Lloyd S. 1953. A Value for n-person Games.Contributions to the Theory of Games. Annals of Mathematical Studies. Princeton University Press.28 (1953), 307–317. doi:10.1515/9781400881970-018

work page doi:10.1515/9781400881970-018 1953
[48]

Oleksandr Shchur, Abdul Fatir Ansari, Caner Turkmen, Lorenzo Stella, Nick Erickson, Pablo Guerron, Michael Bohlke-Schneider, and Yuyang Wang. 2025. fev-bench: A Realistic Benchmark for Time Series Forecasting. doi:10.48550/ ARXIV.2509.26468 Version Number: 1

work page arXiv 2025
[49]

Department of Homeland Security

U.S. Department of Homeland Security. 2024.Roles and responsibilities frame- work for artificial intelligence in critical infrastructure. Technical Report. U.S. Department of Homeland Security. https://www.dhs.gov/sites/default/files/2024- 11/24_1114_dhs_ai-roles-and-responsibilities-framework-508.pdf

2024
[50]

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. InAdvances in Neural Information Processing Systems, Vol. 30. Curran Associates, Inc., 5998–6008

2017
[51]

Gerald Woo, Chenghao Liu, Akshat Kumar, Caiming Xiong, Silvio Savarese, and Doyen Sahoo. 2024. Unified Training of Universal Time Series Forecasting Transformers. doi:10.48550/arXiv.2402.02592 arXiv:2402.02592 [cs]

work page doi:10.48550/arxiv.2402.02592 2024
[52]

Yuyi Zhang, Qiushi Sun, Dongfang Qi, Jing Liu, Ruimin Ma, and Ovanes Pet- rosian. 2024. ShapTime: A General XAI Approach for Explainable Time Series Forecasting. InIntelligent Systems and Applications, Kohei Arai (Ed.). Springer Nature Switzerland, Cham, 659–673. doi:10.1007/978-3-031-47721-8_45

work page doi:10.1007/978-3-031-47721-8_45 2024
[53]

Florian Ziel. 2018. Modeling public holidays in load forecasting: a German case study.Journal of Modern Power Systems and Clean Energy6, 2 (March 2018), 191–207. doi:10.1007/s40565-018-0385-5 A Extended Methodology Chronos-2.Chronos-2 [ 2] is a pretrained foundation model for time series forecasting capable of handling univariate, multivariate, and covari...

work page doi:10.1007/s40565-018-0385-5 2018