ПРОГНОЗИРОВАНИЕ ИНДИКАТОРА СОСТОЯНИЯ РАСТИТЕЛЬНОСТИ НА ОСНОВЕ ГИБРИДНЫХ МОДЕЛЕЙ ГЛУБОКОГО ОБУЧЕНИЯ

Вадим Юрьевич Скобцов; Борис Владимирович Соколов

doi:10.17308/sait/1995-5499/2026/1/135-142

Authors

Vadim Yu. Skobtsov St. Petersburg Federal Research Center of the Russian Academy of Sciences, St.-Petersburg State University of Aerospace Instrumentation
Boris. V. Sokolov St. Petersburg Federal Research Center of the Russian Academy of Sciences

DOI:

https://doi.org/10.17308/sait/1995-5499/2026/1/135-142

Keywords:

predicting, regression, hybrid deep learning models, statistical models, ARIMA model, convolutional one-dimensional layers, recurrent layers, fully connected layers

Abstract

The represented paper considers the problem of forecasting the NDVI (Normalized Difference Vegetation Index) vegetation indicator based on predictive regression analysis of multivariate time series of meteorological data. Various types of hybrid predictive models are developed and evaluated on real statistical data from a subset of the CY-Bench dataset. One-dimensional convolutional neural network, recurrent neural network, and hybrid neural network model based on sequentially connecting blocks of three main types of neural layers (convolutional, recurrent, and fully connected) are developed, evaluated, and tested. It is shown that the proposed hybrid neural network model HNN: 1DCNN + GRU + Dense has the best predicting quality among the studied neural network models on test data according to the metric values RMSE = 0.070 and R^2 = 0.588. A hybrid predictive model combining the linear statistical model ARIMA and a hybrid deep neural network is proposed and evaluated. As a result of applying the proposed hybrid predictive model ARIMA + HNN, the forecasting accuracy for the test dataset was improved to the results of the evaluation indicators RMSE = 0.020 and R^2 = 0.964 with a forecasting horizon of h = 1. With an increase in the forecasting horizon to h = 9, the accuracy level of the improved hybrid model ARIMA + HNN will not drop: RMSE = 0.021 and R^2 = 0.965. Thus, comparative analysis showed that the proposed hybrid model (ARIMA + hybrid neural network) has an advantage in terms of the error value and the determination coefficient compared to the other models under study. It makes further study of hybrid predictive models promising in this area of research.

Author Biographies

Vadim Yu. Skobtsov, St. Petersburg Federal Research Center of the Russian Academy of Sciences, St.-Petersburg State University of Aerospace Instrumentation

PhD in Technical Sciences, Associate Professor
Boris. V. Sokolov, St. Petersburg Federal Research Center of the Russian Academy of Sciences

DSc in Technical Sciences, Professor, Honored Scientist of the Russian Federation, Laureate of the Russian Federation Government Prizes in Science and Technology, Chief Researcher and Head of the Laboratory of Information Technologies in Systems Analysis and Modeling

References

1. CY-Bench: A comprehensive benchmark dataset for subnational crop yield forecasting (1.8) // AgML [Data set] / D. Paudel [et al.]. – 2025. – Режим доступа: https://www.agml.org/. https://doi.org/10.5281/zenodo.15008945.

2. Géron A. Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow / Aurélien Géron. – 2 ed. – Sebastopol, CA: O’Reilly Media, Inc., 2019. – 1065 p.

3. Hybrid Neural Network based on Feature Fusion for Vehicle Type Identification / H. Chen [et al.] // 2020 IEEE International Instrumentation and Measurement Technology Conference (I2MTC), Dubrovnik, Croatia. – 2020. – P. 1–5.

4. Zhao N. Combination of Convolution-al Neural Network and Gated Recurrent Unit for Aspect-Based Sentiment Analysis / N. Zhao // IEEE Access. – Vol. 9. – 2021. – P. 15561–15569.

5. Николенко С. Глубокое обучение / С. Николенко, А. Кадурин, Е. Архангельская. – СПб. : Питер. – 2018. – 480 с.

6. Скобцов В. Ю. Гибридные нейросетевые модели в задаче мультиклассовой классификации данных телеметрической информации малых космических аппаратов / В. Ю. Скобцов, Б. В. Соколов // Вестник ВГУ. Системный анализ и информационные технологии. – 2022. – № 3. – С. 99–114.

7. Скобцов В. Ю. Регрессионный анализ данных сложных систем на основе моделей машинного обучения аппаратов / В. Ю. Скобцов, Б. В. Соколов // Прикладной искусственный интеллект: перспективы и риски: Международная научная конференция (СПб., 17 октября 2024 г.): сб. докл. – СПб. : ГУАП, 2024. – С. 341–344. DOI: 10.31799/978-5-8088- 1975-7-2024

8. Гибридные нейросетевые модели мониторинга данных временных рядов сложных объектов / В. Ю. Скобцов [и др.] // Изв. вузов. Приборостроение. – 2024. – Т. 67, № 2. – С. 200–204. DOI: 10.17586/0021-3454-2024-67- 2-200-204.

9. Brockwell Peter J. Davis Introduction to Time Series and Forecasting / Peter J. Brockwell, Richard A. – New-York : Springer, 2016. – 425 p.

10. Time Series Analysis: Forecasting and Control / George E.P. Box [et al.] – New Jersey: John Wiley & Sons, 2015. – 712 p.