Hourly load forecasting via PatchTST and Temporal Fusion Transformers: comparative system analysis with CatBoost and N-HiTS network models

A.E. Dzgoev, E.V. Klimkin, V.V. Chernyauskas, A.V. Brailovsky, R.N. Rezenkov

---

Abstract. Under modern conditions of dynamic demand behavior and external factors, traditional statistical autoregressive modeling methods (ARIMA, SARIMAX) are increasingly outperformed by machine learning algorithms (Gradient Boosting) and modern deep learning architectures (Transformers).
Aim. This study aims to address the problem of selecting the optimal model for next-day hourly electricity consumption forecasting under limited information constraints, utilizing advanced machine and deep learning methods such as CatBoost, Temporal Fusion Transformer (TFT), PatchTST, and N-HiTS.
Materials and methods. The source code was developed in Python and is publicly available as an open-source repository on GitHub at https://github.com/KEV0143/Comparative-analysis-of-hourly-loadforecasting-using-PatchTST-TFT-NHiTS-and-CatBoost.
Results. New highly accurate and robust mathematical models for electricity consumption forecasting were successfully developed for each of the evaluated methods. A key mathematical outcome of this study demonstrates that effective electricity consumption forecasting for corporate energy conservation can be formulated as a scientifically substantiated selection of the single optimal model. This approach is crucial for both the theoretical analysis of advanced machine learning algorithms and their practical deployment,
driven by the rapidly growing volume of data. Addressing this issue is highly significant for energy consumption optimization across individual enterprises and entire regions.
Conclusion. The theoretical analysis and computational experiments utilizing empirical data from a Russian energy company provided valuable practical insights into the applicability of the CatBoost model for industrial energy consumption optimization. All conclusions presented in this paper are rigorously supported by statistical test results.

Keywords: systems analysis of models, mathematical statistics, short-term load forecasting, electricity consumption, comparative analysis, machine learning, deep learning, time series analysis, CatBoost, Temporal Fusion Transformer (TFT), PatchTST, N‑HiTS

For citation. Dzgoev A.E., Klimkin E.V., Chernyauskas V.V., Brailovsky A.V., Rezenkov R.N. Hourly load forecasting via PatchTST and Temporal Fusion Transformers: comparative system analysis with CatBoost and N-HiTS network models. News of the Kabardino-Balkarian Scientific Center of RAS. 2026. Vol. 28. No. 3. Pp. 49–70. DOI: 10.35330/1991-6639-2026-28-3-49-70

©   Dzgoev A.E., Klimkin E.V., Chernyauskas V.V., Brailovsky A.V., Rezenkov R.N., 2026

References

1. Karpenko S.M., Karpenko N.V., Ematin E.A., Marat Sh. Power consumption planning for an industrial enterprise under the conditions of the wholesale electricity market. Energy Safety and Energy Economy. 2024. No. 6. Pp. 35–40. EDN: CKFPVC. (In Russian)
2. Karpenko S.M., Karpenko N.V., Ematin E.A., Jjunju D. Multiple factor analysis of the wholesale electricity market on the example of a specific pricing zone. Energy Safety and Energy Economy. 2024. No. 4. Pp. 37–42. EDN: QBTQHW. (In Russian)
3. Byk F.L., Myshkina L.S. Economic efficiency of modern electric power industry. Energetik. No. 1. Pp. 17–21. EDN: OGTFOW. (In Russian)
4. Hong T., Fan S. Probabilistic electric load forecasting: A Tutorial Review. International Journal of Forecasting. 2016. Vol. 32. No. 3. Pp. 914–938. DOI: 10.1016/j.ijforecast.2015.11.011
5. Dudek G. Pattern-based local linear regression models for short-term load forecasting. Electric Power System Research. 2016. No. 130. Pp. 139–147. DOI: 10.1016/ j.epsr.2015.09.001
6. Gochhait S., Sharma D.K. Regression model-based short-term load forecasting for load despatch centre. Journal of Applied Engineering and Technological Science. 2023. Vol. 4(2). Pp. 693–710. DOI: 10.37385/jaets.v4i2.1682
7. Lee M.H.L., Ser Y.C., Selvachandran G., Pham T.H. A comparative study of forecasting electricity consumption using machine learning models. Mathematics. 2022. Vol. 10. Article 1329. DOI: 10.3390/math10081329
8. Dayarathna M., Wen Yo., Fan Rui. Data center energy consumption modeling: a survey. IEEE Communications Surveys & Tutorials. 2016. Vol. 18. No. 1. DOI: 10.1109/COMST.2015.2481183
9. Mughees M., Li Y., Chen Y., Li Y.R. Short-term load forecasting for ai-data center. IEEE PES General Meeting 2025. Electrical Engineering and Systems Science. 2025. DOI: 10.48550/arXiv.2503.07756
10. Ye Z., Gao W., Hu Q. et al. Deep learning workload scheduling in GPU datacenters: A survey. ACM Computing Surveys. 2024. Vol. 56. No. 6. Pp. 1–38. DOI: 10.1145/3638757
11. Munkhammar J., Meer D., Widén J. Very short term load forecasting of residential electricity consumption using the Markov-chain mixture distribution (MCM) model. Applied Energy. 282(A):116180. DOI: 10.1016/j.apenergy.2020.116180
12. Hippert H.S., Pedreira C.E., Souza R.C. Neural networks for short-term load forecasting: a review and evaluation. IEEE Transactions on Power Systems. Vol. 16. No. 1. Pp. 44–55. DOI: 10.1109/59.910780
13. Sailor D.J., Munoz J.R. Sensitivity of electricity and natural gas consumption to climate in the U.S.A. Methodology and results for eight states. Energy. 1997. Vol. 22. No. 10. Pp. 987–998. DOI: 10.1016/S0360-5442(97)00034-0
14. Hong T., Pinson P., Fan S. et al. Probabilistic energy forecasting: global energy forecasting competition 2014 and beyond. International Journal of Forecasting. 2016. Vol. 32. No. 3. Pp. 896–913. DOI: 10.1016/j.ijforecast.2016.02.001
15. Wang A., Yu Q., Wang J. et al. Electric load forecasting based on deep ensemble learning. Applied Sciences. 2023. Vol. 13. No. 17. P. 9706. DOI: 10.3390/app13179706
16. Lim B., Arık Ö.S., Loeff N., Pfister T. Temporal fusion transformer for interpretable multi-horizon time series forecasting. International Journal of Forecasting. 2021. Vol. 37. No. 1. DOI: 10.1016/j.ijforecast.2021.03.012
17. Challu C., Olivares K.G., Oreshkin B.N. et al. N‑HiTS: Neural hierarchical interpolation for time series forecasting. Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 37. No. 6. Pp. 6989–6997. DOI: 10.48550/arXiv.2201.12886
18. Ahmad H., Mortazavi S.K., Bahnasawi M.E. et al. Enhanced time series forecasting: integrating PatchTST with BERT Layers. Conference: 5th International Conference on Applied
    Mathematics & Computer Science (ICAMCS 2025). At: Venice, Italy, September 27–29, 2025. DOI: 10.1109/ICAMCS62774.2024.00014
19. Dorogush A.V., Ershov V., Gulin A. Catboost: gradient boosting with categorical features support. DOI: 10.48550/arXiv.1810.11363.2018
20. Lim B., Arık S.Ö., Loeff N., Pfister T. Temporal fusion transformer for interpretable multi-horizon time series forecasting. International Journal of Forecasting. Vol. 37. No. 4. Pp. 1748–1764. DOI: 10.1016/j.ijforecast.2021.03.012
21. Nie Y., Nguyen N.H., Sinthong P., Kalagnanam J. A time series is worth 64 words: long-term forecasting with transformers (PatchTST). Proceedings of the International Conference on Learning Representations. International Conference on Learning Representations ICLR 2023. https://arxiv.org/pdf/2211.14730.pdf

Information about the authors

Alan E. Dzgoev, Candidate of Technical Sciences, Associate Professor, Associate Professor at the Department of Digital Transformation, Institute of Information Technologies, MIREA – Russian Technological University;
78, Vernadsky prospekt, Moscow, 119454, Russia;
dzgoev@mirea.ru, ORCID: https://orcid.org/0000-0002-1314-6151, SPIN-code: 8092-8784
Egor V. Klimkin, Undergraduate Student, Department of Applied Mathematics, Institute of Information Technologies, MIREA – Russian Technological University;
78, Vernadsky prospekt, Moscow, 119454, Russia.;
KlimkinEVK@yandex.ru, ORCID: https://orcid.org/0009-0001-3876-3041
Vladislav V. Chernyauskas, Senior Lecturer, Department of Digital Transformation, Institute of Information Technologies, MIREA – Russian Technological University;
78, Vernadsky prospekt, Moscow, 119454, Russia;
chernyauskas-vladislav@yandex.ru, ORCID: https://orcid.org/0009-0002-8438-3418, SPIN-code: 5867-6366
Andrey V. Brailovsky, Assistant Professor, Department of Digital Transformation, Institute of Information Technologies, MIREA – Russian Technological University;
78, Vernadsky prospekt, Moscow, 119454, Russia;
brajlovskij@mirea.ru, ORCID: https://orcid.org/0009-0006-1794-7825, SPIN-code: 5900-1835
Roman N. Rezenkov, Candidate of Technical Sciences, Associate Professor, Associate Professor at the Digital Transformation Department, Institute of Information Technologies, MIREA – Russian Technological University;
78, Vernadsky prospekt, Moscow, 119454, Russia;
rezenkov@mirea.ru, ORCID: https://orcid.org/0009-0005-5542-2125

Funding

The study was performed without external funding.