ellibsЭлектронные библиотекиRussian Digital Libraries Journal1562-5419Казанский (Приволжский) федеральный университет10.26907/1562-5419-2026-29-1-351-367ellibs-731Research ArticleСтатьиМульти-таймфреймовые Drummond-патчи и JEPA-предобучение для краткосрочного прогноза розничных OHLC-рядовMulti-Timeframe Drummond Patches and JEPA Pre-Training for Short-Term Retail OHLC Series ForecastingСизовАлександр СемёновичSizovAlexander Semenovichkafedra-ipm@mail.ruХалинЮрий АлексеевичKhalinYuri Alekseevichyur-khalin@yandex.ruБелыхАртём АлександровичBelykhArtem Aleksandrovichbelykhartem.a@mail.ruЮго-Западный государственный университетSouthwest State University202604032026291351367Copyright © Сизов А.С., Халин Ю.А., Белых А.А., 20262026Сизов А.С., Халин Ю.А., Белых А.А.Sizov A.S., Khalin Y.A., Belykh A.A.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://ellibs.elpub.ru/jour/article/view/731

Предложен метод построения инвариантных к масштабу представлений временных рядов розничной выручки на базе трехбарной (по трем соседним периодам) геометрии Драммонда (DG), расширенной мульти-таймфреймовым контекстом (день, частичная календарная неделя и скользящая 7-дневка). На этих «патчах» выполнено self-supervised предобучение по схеме Joint-Embedding Predictive Architecture (JEPA) со спатио-темпоральным маскированием, после чего модель дообучена с выходными слоями, оценивающими неопределенность, для прогноза на следующий день и следующую неделю. Проанализированы свойства аффинной инвариантности признаков и идентифицируемости недельной фазы; эмпирически продемонстрировано улучшение по сравнению с сильными базовыми моделями на реальных данных.

We propose a method for constructing scale-invariant representations of retail revenue time series based on three-bar Drummond Geometry (DG) computed over three adjacent periods, extended with a multi-timeframe context (day, partial calendar week, and a rolling 7-day window). Self-supervised pre-training on these “patches” is performed using a Joint-Embedding Predictive Architecture (JEPA) with spatiotemporal masking, followed by fine-tuning with output heads that quantify predictive uncertainty for next-day and next-week forecasts. The work analyzes the properties of affine invariance of the features and the identifiability of the weekly phase; empirical improvement over strong baseline models on real-world data is demonstrated.

геометрия ДраммондаJoint-Embedding Predictive Architecture (JEPA)временные рядыOpen-High-Low-Close (OHLC)розничная торговлякраткосрочный прогнозсамообучениеDrummond GeometryJoint-Embedding Predictive Architecture (JEPA)time seriesOpen-High-Low-Close (OHLC)retailshort-term forecastingself-supervised learningReferencesFildes R., Ma S., Kolassa S. Retail forecasting: Research and practice // International Journal of Forecasting. 2022. Vol. 38, No. 4. P. 1283–1318. https://doi.org/10.1016/j.ijforecast.2019.06.004Fildes R., Ma S., Kolassa S. Retail forecasting: Research and practice // International Journal of Forecasting. 2022. Vol. 38, No. 4. P. 1283–1318. https://doi.org/10.1016/j.ijforecast.2019.06.004Lim B., Arik S.O., Loeff N., Pfister T. Temporal Fusion Transformers for interpretable multi-horizon time series forecasting // International Journal of Forecasting. 2021. Vol. 37, No. 4. P. 1748–1764. https://doi.org/10.1016/j.ijforecast.2021.03.012Lim B., Arik S.O., Loeff N., Pfister T. Temporal Fusion Transformers for interpretable multi-horizon time series forecasting // International Journal of Forecasting. 2021. Vol. 37, No. 4. P. 1748–1764. https://doi.org/10.1016/j.ijforecast.2021.03.012Hyndman R.J., Athanasopoulos G. Forecasting: Principles and Practice. 2nd ed. Melbourne: OTexts, 2018. 380 p. Cited pp.: 183–220, 221–274, 347–368.Hyndman R.J., Athanasopoulos G. Forecasting: Principles and Practice. 2nd ed. Melbourne: OTexts, 2018. 380 p. Cited pp.: 183–220, 221–274, 347–368.Oreshkin B.N., Carpov D., Chapados N., Bengio Y. N-BEATS: Neural basis expansion analysis for interpretable time series forecasting // arXiv preprint arXiv:1905.10437. 2019. https://doi.org/10.48550/arXiv.1905.10437; Challu C. et al. NHITS: Neural hierarchical interpolation for time series forecasting // Proceedings of the AAAI Conference on Artificial Intelligence. 2023. Vol. 37, No. 6. P. 6989–6997. https://doi.org/10.1609/aaai.v37i6.25854Oreshkin B.N., Carpov D., Chapados N., Bengio Y. N-BEATS: Neural basis expansion analysis for interpretable time series forecasting // arXiv preprint arXiv:1905.10437. 2019. https://doi.org/10.48550/arXiv.1905.10437; Challu C. et al. NHITS: Neural hierarchical interpolation for time series forecasting // Proceedings of the AAAI Conference on Artificial Intelligence. 2023. Vol. 37, No. 6. P. 6989–6997. https://doi.org/10.1609/aaai.v37i6.25854Yue Zh. et al. TS2Vec: Towards Universal Representation of Time Series // Proceedings of the AAAI Conference on Artificial Intelligence. 2022. Vol. 36, No. 8. P. 8980–8987. https://doi.org/10.1609/aaai.v36i8.20881Yue Zh. et al. TS2Vec: Towards Universal Representation of Time Series // Proceedings of the AAAI Conference on Artificial Intelligence. 2022. Vol. 36, No. 8. P. 8980–8987. https://doi.org/10.1609/aaai.v36i8.20881Hearne T. Drummond Geometry: Picking Yearly Highs and Lows in Interbank Forex Trading // Breakthroughs in Technical Analysis: New Thinking from the World’s Top Minds / ed. by D. Keller. Princeton: Bloomberg Press, 2007. P. 1–19. https://doi.org/10.1002/9781119204749.ch1Hearne T. Drummond Geometry: Picking Yearly Highs and Lows in Interbank Forex Trading // Breakthroughs in Technical Analysis: New Thinking from the World’s Top Minds / ed. by D. Keller. Princeton: Bloomberg Press, 2007. P. 1–19. https://doi.org/10.1002/9781119204749.ch1Dawid A., LeCun Y. Introduction to Latent Variable Energy-Based Models: A Path Towards Autonomous Machine Intelligence // Journal of Statistical Mechanics: Theory and Experiment. 2024. No. 10. Art. 104011. https://doi.org/10.1088/1742-5468/ad292b (arXiv:2306.02572).Dawid A., LeCun Y. Introduction to Latent Variable Energy-Based Models: A Path Towards Autonomous Machine Intelligence // Journal of Statistical Mechanics: Theory and Experiment. 2024. No. 10. Art. 104011. https://doi.org/10.1088/1742-5468/ad292b (arXiv:2306.02572).Assran M. et al. Self-supervised learning from images with a joint-embedding predictive architecture // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2023. P. 15619–15629. https://doi.org/10.1109/CVPR52729.2023.01499Assran M. et al. Self-supervised learning from images with a joint-embedding predictive architecture // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2023. P. 15619–15629. https://doi.org/10.1109/CVPR52729.2023.01499Park Y.J. et al. A scalable and transferable time series prediction framework for demand forecasting. 2024. arXiv preprint arXiv:2402.19402. https://doi.org/10.48550/arXiv.2402.19402Park Y.J. et al. A scalable and transferable time series prediction framework for demand forecasting. 2024. arXiv preprint arXiv:2402.19402. https://doi.org/10.48550/arXiv.2402.19402Ragab M., Liu Q., Jia W., Chen M., Yun U. Self-Supervised Learning for Time Series Analysis: Taxonomy, Progress, and Prospects // IEEE Transactions on Pattern Analysis and Machine Intelligence. 2024. Vol. 46, No. 10. P. 6775–6794. https://doi.org/10.1109/TPAMI.2024.3387317Ragab M., Liu Q., Jia W., Chen M., Yun U. Self-Supervised Learning for Time Series Analysis: Taxonomy, Progress, and Prospects // IEEE Transactions on Pattern Analysis and Machine Intelligence. 2024. Vol. 46, No. 10. P. 6775–6794. https://doi.org/10.1109/TPAMI.2024.3387317Voloshin T.A., Zaitsev K.S., Dunaev M.E. Application of adaptive ensembles of machine learning methods to the problem of time series forecasting // International Journal of Open Information Technologies. 2023. Vol. 11, No. 8. P. 57–63.Voloshin T.A., Zaitsev K.S., Dunaev M.E. Application of adaptive ensembles of machine learning methods to the problem of time series forecasting // International Journal of Open Information Technologies. 2023. Vol. 11, No. 8. P. 57–63.Diebold F.X., Mariano R.S. Comparing Predictive Accuracy // Journal of Business & Economic Statistics. 1995. Vol. 13, No. 3. P. 253–263. https://doi.org/10.1080/07350015.1995.10524599Diebold F.X., Mariano R.S. Comparing Predictive Accuracy // Journal of Business & Economic Statistics. 1995. Vol. 13, No. 3. P. 253–263. https://doi.org/10.1080/07350015.1995.10524599

The authors declare that there are no conflicts of interest present.