A.V. Gulyaev1
1 Russian State Social University (Moscow, Russia)

1 Gulyaev81@gmail.com

Time series forecasting plays a pivotal role in various fields, such as economics, meteorology, energy, and others. This paper presents the architecture of a hybrid deep learning model designed for forecasting time series values. Traditional models, developed based on machine learning methods and statistical approaches, demonstrate certain achievements and successes in addressing a broad range of tasks across various applied fields. However, despite their potential strength and versatility, these models face several limitations in situations where data exhibit high complexity and unpredictability. In conditions where data are characterized by high dimensionality, unstructured nature, or contain a significant amount of noise and anomalies, traditional approaches may lose the ability to adequately interpret and extract useful information from such data. To overcome these limitations, a hybrid model combining the Prophet forecasting model, recurrent neural networks (RNNs), and convolutional neural networks (CNNs) has been proposed. Convolutional neural networks have the capability to detect patterns that are inaccessible to statistical methods, while recurrent neural networks are effective for processing sequential data. Thus, combination of these architectures allows for improved forecasting quality. Experiments conducted on various datasets have shown that the proposed hybrid model outperforms some conventional statistical methods and individual machine learning models in terms of forecast accuracy. The developed model is promising for applications requiring a high level of predictive accuracy in conditions of complex temporal dependencies.

Gulyaev A.V. Hybrid deep learning model for time series forecasting. Neurocomputers. 2025. V. 27. № 3. P. 62–72. DOI: https://doi.org/ 10.18127/j19998554-202503-08 (in Russian)

1. Gulyaev A.V., Pivneva S.V. Analiz sushchestvuyushchikh regressionnykh, avtoregressionnykh modelej, modelej eksponentsial'nogo sglazhivaniya. Ikh primenenie k prognozirovaniyu pokupatel'skogo sprosa produktov pitaniya, kharakteristiki, osobennosti, dostoinstva i nedostatki. Sb. nauch. trudov Vseros. nauch.-tekhnich. konf. «Nauka. Proizvodstvo. Obrazovanie». M.: Rossijskij gosudarstvennyj sotsial'nyj universitet. 2023. S. 41–51. (in Russian)
2. Orlova I.V. Ispol'zovanie paketa Prophet v prognozirovanii vremennykh ryadov. Fundamental'nye issledovaniya. 2021. № 3. S. 94–102. DOI: 10.17513/fr.42987. (in Russian)
3. Gulyaev A.V. Primenenie vejvlet-preobrazovaniya i singulyarnogo spektral'nogo analiza pri dekompozitsii vremennogo ryada. Sistemy vysokoj dostupnosti. 2024. № 2. S. 76–84. DOI: 10.18127/j20729472-202402-06. (in Russian)
4. Porshnev S.V., Rabaja F. Issledovanie osobennostej primeneniya metoda singulyarnogo spektral'nogo analiza v zadache analiza i prognozirovaniya vremennykh ryadov: Monografiya. Ul'yanovsk: Zebra. 2016. (in Russian)
5. Vil'danov N.R., Emaletdinova L.Yu. Obzor metodov prognozirovaniya elementov vremennogo ryada. Sb. materialov X Mezhdunar. nauch.-praktich. konf. «Razvitie nauki i praktiki v global'no menyayushchemsya mire v usloviyakh riskov». M.: OOO «IROK». 2022. S. 114–119. DOI: 10.34755/IROK.2022.65.53.022. (in Russian)
6. Mahjoub S., Chrifi-Alaoui L., Marhic B. et al. Predicting energy consumption using LSTM, multi-layer GRU and drop-GRU neural networks. Sensors. 2022. V. 22. № 11. P. 4062. DOI: 10.3390/s22114062.
7. Arulmurugan A. et al. A model for predicting E-Commerce product returns using hybrid CNN-GRU. International Journal on Recent and Innovation Trends in Computing and Communication. 2023. V. 11. № 9. P. 3615–3619. DOI: 10.17762/ijritcc.v11i9.9583.
8. Jang J., Han J., Leigh S.B. Prediction of heating energy consumption with operation pattern variables for non-residential buildings using LSTM networks. Energy and Buildings. 2022. V. 255. P. 111647. DOI: 10.1016/j.enbuild.2021.111647.
9. Mandal U., Chakraborty A., Mahato Ph. et al. Bogaba: a novel hybrid deep learning model for time series forecasting using sentiment analysis. Indian Journal of Computer Science and Engineering. 2023. V. 14. № 1. P. 130–145. DOI: 10.21817/indjcse/2023/v14i1/ 231401121.
10. Nagdiya A., Kapoor V., Tokekar V. A hybrid deep learning model for accurate time series forecasting of cryptocurrencies. Journal of Information and Optimization Sciences. 2024. V. 45. № 4. P. 1061–1072. DOI: 10.47974/jios-1691.