Сomparative evaluation of machine and deep learning models for short-term electricity consumption forecasting
DOI:
https://doi.org/10.33216/1998-7927-2026-304-6-161-169Keywords:
Internet of Things, IoT, energy forecasting, artificial intelligence, machine learning, deep learning, LSTMAbstract
The growth of digitalization of energy systems, the development of Internet of Things (IoT) technologies and the implementation of the Smart Grid concept necessitate increasing the accuracy of short-term electricity consumption forecasting. The efficiency of electrical load management, optimization of energy storage systems, integration of renewable energy sources and decision support in intelligent energy systems depend on the quality of the forecast. At the same time, there are no unambiguous conclusions regarding the advantages of complex deep learning models over modern machine learning algorithms for forecasting regular time series of electricity consumption.
The aim of the work is to comparatively assess the effectiveness of machine and deep learning models for short-term energy consumption forecasting, as well as to analyze the impact of the temporal attention mechanism (Attention) on the forecasting quality of a recurrent neural network of the Long Short-Term Memory (LSTM) type.
To ensure the reproducibility of the results, the open dataset Power Consumption of Tetouan City (UCI Machine Learning Repository) was used, which contains over 52 thousand electricity consumption measurements and synchronized meteorological parameters. Data aggregation to hourly discretization, feature standardization, cyclic encoding of calendar characteristics and formation of training sequences using a sliding time window of 24 hours were performed. Within the framework of a single experimental methodology, a comparative study of the Persistence, Ridge Regression, Random Forest, HistGradientBoosting, LSTM, GRU and LSTM-Attention models was conducted. The quality of prediction was assessed by MAE, RMSE, MAPE and coefficient of determination (R2).
The experimental results showed that the best prediction quality for the studied data set was provided by the Ridge Regression and HistGradientBoosting models, which achieved RMSE values of 898.29 and 901.13, respectively, a coefficient of determination of more than 0.977 and a MAPE of about 2.2%. It was found that the application of the temporal attention mechanism allowed to improve the results of the recurrent LSTM model: the mean square error decreased by 16.70%, and the mean absolute error by 23.66% compared to the standard LSTM. However, for the studied time series, the proposed LSTM-Attention model did not outperform the best regression models, which indicates the feasibility of choosing a forecasting algorithm depending on the properties of a specific data set. The obtained results confirm the effectiveness of using the Attention mechanism to improve the quality of recurrent neural networks and demonstrate that for regular hourly time series, modern regression algorithms can provide higher forecasting accuracy than complex deep learning models. The practical significance of the study lies in the possibility of using the obtained results when choosing forecasting models for energy management systems, smart grids, and analytical platforms for processing energy data.
References
1. Gubbi J., Buyya R., Marusic S., Palaniswami M. Internet of Things (IoT): A vision, architectural elements, and future directions. Future Generation Computer Systems. 2013. Vol. 29. No. 7. P. 1645–1660. DOI: https://doi.org/10.1016/j.future.2013.01.010
2. Dzheria T., Shevchuk V. Forecasting electricity consumption using neural networks. Control, Navigation and Communication Systems. 2023. No. 2(72). P. 42–44. DOI: https://doi.org/10.26906/SUNZ.2023.2.042
3. Крилова Т. В., Худа Ж. В., Балясний О. М. Прогнозування обсягів електроспоживання з використанням нейромережевих технологій. Математичне моделювання. 2023. № 2(49). С. 116–124. DOI: https://doi.org/10.31319/2519-8106.2(49)2023.292610
4. Шиманюк П. В., Мірошник В. О., Блінов І. В. Визначення втрат електричної енергії на основі прогнозів вузлового електричного навантаження. Енергетика: економіка, технології, екологія. 2022. № 3. С. 83–91. DOI: https://doi.org/10.20535/1813-5420.3.2022.271484
5. Лободзинський В., Бурик М., Паламарчук А. Прогнозування попиту на електроенергію в Smart Grids на основі моделей машинного навчання. Енергетика: економіка, технології, екологія. 2026. № 1. DOI: https://doi.org/10.20535/1813-5420.1.2026.351525
6. Лоскутов С. С., Шиманюк П. В. Прогнозування електричного навантаження на ієрархічних рівнях ОЕС України з використанням нейронної мережі типу LSTM. Праці Інституту електродинаміки НАН України. 2021. № 59. С. 69–75. DOI: https://doi.org/10.15407/publishing2021.59.081
7. Yan K., Zhou X., Chen J. Collaborative deep learning framework on IoT data with bidirectional NLSTM neural networks for energy consumption forecasting. Journal of Parallel and Distributed Computing. 2022. Vol. 163. P. 248–255. DOI: https://doi.org/10.1016/j.jpdc.2022.01.012
8. Mounir N., Ouadi H., Jrhilifa I. Short-term electric load forecasting using an EMD-BI-LSTM approach for smart grid energy management system. Energy and Buildings. 2023. Vol. 288. Article 113022. DOI: https://doi.org/10.1016/j.enbuild.2023.113022
9. Sideratos G., Paspatis A., Kotsampopoulos P. et al. Deployment of forecasting tools in diverse demonstration areas to improve energy scheduling of microgrids. In: Proceedings of the 27th International Conference on Electricity Distribution (CIRED). Rome, 2023. DOI: https://doi.org/10.1049/icp.2023.1199
10. Sinitsyn I. P., Doroshenko A. Yu., Yatsenko O. A. et al. Research of software solutions for forecasting electricity generation and consumption in Ukraine that are based on machine learning methods. Проблеми програмування. 2023. № 3. С. 99–113. DOI: https://doi.org/10.15407/pp2023.03.099
11. Salam A., El Hibaoui A. Comparison of Machine Learning Algorithms for the Power Consumption Prediction: Case Study of Tetouan City // 2018 6th International Renewable and Sustainable Energy Conference (IRSEC). IEEE, 2018. P. 1–5. DOI: https://doi.org/10.1109/IRSEC.2018.8703007
12. Salam A., El Hibaoui A. Power Consumption of Tetouan City [Dataset]. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5B034
13. Hochreiter S., Schmidhuber J. Long Short-Term Memory. Neural Computation. 1997. Vol. 9. No. 8. P. 1735–1780.
14. Vaswani A., Shazeer N., Parmar N. et al. Attention Is All You Need. In: Advances in Neural Information Processing Systems. 2017. Vol. 30. P. 5998–6008.
15. Gonzalez Grandon T., Schwenzer J., Steens T., Breuing J. Electricity Demand Forecasting with Hybrid Statistical and Machine Learning Algorithms: Case Study of Ukraine. Applied Energy. 2024. Vol. 355. Article 122295. DOI: https://doi.org/10.1016/j.apenergy.2023.122249
16. Han J., Kamber M., Pei J. Data Mining: Concepts and Techniques. 4th ed. Cambridge, MA: Morgan Kaufmann, 2023.
