Forecasting time series of technological parameters of equipment using LSTM recurrent neural networks

The digitalization of the agro-industrial sector and the growing volume of data generated by agricultural machinery and technological equipment create a demand for intelligent analytical methods capable of capturing nonlinear dynamics and adapting to changing operating modes. Time-series forecasting methods for technological parameters are of particular interest, as they can improve monitoring accuracy and enable the transition to predictive control and maintenance strategies. (Research purpose) The study aims to assess the potential of Long Short-Term Memory (LSTM) recurrent neural networks for short-term forecasting time series of technological parameters in agricultural machinery and equipment, taking into account long-term temporal dependencies and the non-stationary nature of the data. (Materials and methods) The study used telemetric time series obtained during the operation of industrial equipment as the initial data. Training sequences were developed to preserve the temporal context of the observations. Modelling was performed using a recurrent neural network based on the LSTM architecture, which allows long-term dependencies in the data to be captured. Forecasting performance was evaluated using the MAE, MSE, and RMSE metrics. ARIMA models and a fully connected neural network were employed as baseline methods for comparison. (Results and discussion) The developed LSTM model demonstrated high forecasting accuracy, effectively reproducing both stationary segments and transition periods of the time series. The experimental results yielded MAE = 0.0094, MSE = 0.00014, and RMSE = 0.0119, indicating that the proposed LSTM model outperformed the benchmark models. (Conclusions) The results confirm the effectiveness of LSTM-based models for the analysis and forecasting of technological processes and indicate their strong potential for application in industrial monitoring and predictive control systems.

1. Kostomakhin M.N., Pestryakov E.V. Software package for remote diagnostics of agricultural machinery condition. Agricultural Machinery and Technologies. 2022. Vol. 16. N4. 19-25 (In Russian). DOI: 10.22314/2073-7599-2022-16-4-19-25.

2. Pashshoev B., Petrusevich D.A. Neural network analysis in time series forecasting. Russ. Technol. J. 2024;12(4):106−116. DOI: 10.32362/2500-316X-2024-12-4-106-116.

3. Qin S.J. Process data analytics in the era of big data. AIChE Journal. 2020. Vol. 66. Iss. 4. e16865. DOI:10.1002/aic.14523.

4. Napoleone A., Macchi M., Pozzetti A. A review on the characteristics of cyber-physical systems for future smart factories. Journal of Manufacturing Systems. 2020. Vol. 54. 305-335. DOI: 10.1016/j.jmsy.2020.01.007.

5. Gao Y., Wang R., Zhou E. Stock prediction based on optimized LSTM and GRU models. Wireless Communications and Mobile Computing. 2021. 4055281. DOI: 10.1155/2021/4055281.

6. Mathonsi T., van Zyl T.L. A statistics and deep learning hybrid method for multivariate time series forecasting and mortality modeling. Forecasting. 2022. Vol. 4. N1. 1-25. DOI: 10.48550/arXiv.2112.08618.

7. Zhang C., Sjarif N.N.A., Ibrahim R. Deep learning models for price forecasting of financial time series: A review of recent advancements: 2020-2022. 2023. DOI: 10.1002/widm.1519.

8. Smyl S., Dudek G., Pelka P. ES-dRNN: a hybrid exponential smoothing and dilated recurrent neural network model for short-term load forecasting. 2112.02663 [cs.LG]. 2021. DOI: 10.48550/arXiv.2112.02663.

9. Savvin N.V. Time series forecasting: improving LSTM models using vector-time encoding. Engineering Systems and Structures. 2025. N2 (60). 148-154 (In Russian). DOI: 10.36622/2074-188X.2025.54.65.015.

10. Zhang G., Ren T., Yang Y. A new unified deep learning approach with decomposition-reconstruction-ensemble framework for time series forecasting. 2002.09695 [stat. ML]. 2020. DOI: 10.48550/arXiv.2002.09695.

11. Erichson N.B., Lim S.H., Mahoney M. W. Gated recurrent neural networks with weighted time-delay feedback. 2212.00228 [cs.LG]. 2022. DOI: 10.48550/arXiv.2212.00228.

12. Pashshoev B.A., Petrusevich D.A. Analysis of neural network models for time series forecasting. Russian Technological Journal. 2024. Vol. 12. N4. 106-116. (In Russian). DOI: 10.32362/2500-316X-2024-12-4-106-116.

13. Nurhaida I., Noprisson H., Ayumi V. et al. Implementation of deep learning predictor (LSTM) algorithm for human mobility prediction. 2020. International Journal of Interactive Mobile Technologies. 2020. Vol. 14. N18. 132-144. DOI: 10.3991/ijim.v14i18.16867.

14. Rapakov G.G., Gorbunov V.A., Dianov S.V., Elizarova L.V. Study of LSTM neural network approach in time series modeling. Bulletin of Cherepovets State University. 2023. N3 (114). 47-54 (In Russian). DOI: 10.23859/1994-0637-2023-3-114-4.

15. Wang Q., Chen L., Zhao J., Wang W. A deep granular network with adaptive unequal-length granulation strategy for long-term time series forecasting and its industrial applications. Artificial Intelligence Review. 2020. Vol. 53. N7. 5353-5381. DOI: 10.1007/s10462-020-09822-9.

16. Smirnov I.G., Khort D.O., Kutyrev A.I. Intelligent technologies and robotic machines for garden crops cultivation. Agricultural Machinery and Technologies. 2021. Vol. 15. N4. 35-41 (In Russian). DOI: 10.22314/2073-7599-2021-15-4-35-41.