Comparative Forecasting of Electrical Load Demand Using ANN, SVR, and RFR Models: An AI-Based Approach

Abstract

Accurate forecasting of electrical load demand is essential for efficient energy management and planning. This study presents a comparative analysis of three artificial intelligence models—Artificial Neural Network (ANN), Support Vector Regression (SVR), and Random Forest Regression (RFR)—for short-term load forecasting. The models were trained using historical load data from POSOCO and weather parameters such as temperature and humidity obtained from NASA POWER and OpenWeatherMap. Performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R² score. Results show that the RFR model outperformed the others, achieving the highest accuracy with an R² value of 0.93. The ANN model also performed well, while SVR showed slightly lower predictive accuracy. The study highlights the potential of ensemble learning for improving load forecasting reliability.

Introduction

The increasing demand for electricity, driven by urbanization and digitization, has made accurate short-term load forecasting (STLF) essential for efficient power system operations. Traditional statistical methods like ARIMA and exponential smoothing struggle to handle the nonlinear, stochastic nature of modern electricity demand influenced by weather, temporal, and socio-economic factors.

To address this, modern forecasting has shifted to AI and machine learning (ML) models. The study compares three prominent AI models:

• Artificial Neural Networks (ANN)

• Support Vector Regression (SVR)

• Random Forest Regression (RFR)

Key Concepts and Models

• ANN: Mimics brain-like learning to model complex nonlinear relationships but may require large data and careful tuning.

• SVR: Works well with smaller datasets using kernel tricks, but is sensitive to parameter tuning.

• RFR: An ensemble method that aggregates multiple decision trees, offering high accuracy and robustness.

Research Gap & Objective

Most past studies focus on individual or hybrid models without fair comparisons under uniform conditions. This study fills that gap by evaluating ANN, SVR, and RFR on the same dataset, using the same input features and performance metrics.

Methodology

• Data Source: Hourly load data from POSOCO and weather data from NASA POWER and OpenWeatherMap.

• Input Features: Temperature, humidity, past load, time of day, day of week, holiday indicator, and season/month.

• Model Configurations:

  • ANN: 3 hidden layers, ReLU activation, Adam optimizer.

  • SVR: RBF kernel, optimized parameters.

  • RFR: 300 trees, max depth 20.

• Evaluation Metrics: MAE, RMSE, R².

• Tools Used: Python (Scikit-learn, Keras, TensorFlow, Pandas, etc.).

Results

• RFR emerged as the best-performing model, balancing accuracy, robustness, and generalizability.

• ANN performed well but was slightly less accurate.

• SVR lagged in handling complex, nonlinear dependencies.

Conclusion

This study aimed to evaluate the effectiveness of three widely used artificial intelligence techniques—Artificial Neural Network (ANN), Support Vector Regression (SVR), and Random Forest Regression (RFR)—for short-term electrical load forecasting. Using a robust dataset composed of hourly load demand from POSOCO, and weather parameters such as temperature and humidity obtained from NASA POWER and OpenWeatherMap, the models were trained and tested under consistent conditions. Additional calendar variables such as time of day, day of the week, and holiday indicators were included to reflect human activity patterns that influence energy usage. The findings highlight the superiority of the Random Forest Regression model, which outperformed both ANN and SVR in terms of accuracy and consistency. With the lowest MAE (17.3 MW), RMSE (23.1 MW), and highest R² value (0.93), RFR demonstrated strong capability in capturing non-linear and noisy trends in the dataset. While the ANN model also achieved high accuracy (R² = 0.91), it required more computational resources and parameter tuning. The SVR model, though suitable for smaller datasets, yielded comparatively higher errors and was less effective in modeling the complexity of real-world load patterns. These results offer valuable insights for utility operators and energy planners, emphasizing the practical viability of ensemble learning approaches such as RFR in developing reliable load forecasting systems. Accurate short-term forecasts not only aid in optimizing power generation and reducing operational costs but also enhance demand response and grid reliability, particularly in dynamic and resource-constrained settings like India. Looking ahead, there are several promising directions for future research. The integration of advanced deep learning models such as Long Short-Term Memory (LSTM) or Gated Recurrent Units (GRU) could help capture sequential dependencies in load data more effectively.

References

[1] Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324 [2] Chen, K., Chen, J., Wang, Q., & Wang, Z. (2020). Short-term load forecasting using artificial intelligence. Energy Reports, 6, 138–145. https://doi.org/10.1016/j.egyr.2019.08.041 [3] Haykin, S. (2009). Neural networks and learning machines (3rd ed.). Pearson. [4] Hippert, H. S., Pedreira, C. E., & Souza, R. C. (2001). Neural networks for short-term load forecasting: A review and evaluation. IEEE Transactions on Power Systems, 16(1), 44–55. https://doi.org/10.1109/59.910780 [5] Hong, T., & Fan, S. (2016). Probabilistic electric load forecasting: A tutorial review. International Journal of Forecasting, 32(3), 914–938. https://doi.org/10.1016/j.ijforecast.2015.11.011 [6] Hyndman, R. J., & Athanasopoulos, G. (2018). Forecasting: Principles and practice (2nd ed.). OTexts. https://otexts.com/fpp2/ [7] Khosravi, A., Nahavandi, S., Creighton, D., & Atiya, A. F. (2018). Comprehensive review of neural network-based short-term electricity load forecasting. Energy Systems, 9(2), 107–132. https://doi.org/10.1007/s12667-017-0256-2 [8] Kuster, C., Rezgui, Y., &Mourshed, M. (2017). Electrical load forecasting models: A critical systematic review. Sustainable Cities and Society, 35, 257–270. https://doi.org/10.1016/j.scs.2017.08.009 [9] Li, K., Su, H., & Shu, L. (2019). Load forecasting via hybrid machine learning model integrating empirical mode decomposition and random forest. Applied Energy, 237, 147–157. https://doi.org/10.1016/j.apenergy.2019.01.031 [10] Pai, P. F., & Hong, W. C. (2005). Support vector machines with simulated annealing algorithms in electricity load forecasting. Energy Conversion and Management, 46(17), 2669–2688. https://doi.org/10.1016/j.enconman.2004.12.007 [11] Park, D. C., El-Sharkawi, M. A., Marks II, R. J., Atlas, L. E., & Damborg, M. J. (1991). Electric load forecasting using an artificial neural network. IEEE Transactions on Power Systems, 6(2), 442–449. https://doi.org/10.1109/59.76868 [12] Rahman, S., & Bhatnagar, S. (2019). Hour-ahead electrical load forecasting using ANN model for utility operations. Energy Procedia, 158, 2568–2573. https://doi.org/10.1016/j.egypro.2019.01.092 [13] Smola, A. J., &Schölkopf, B. (2004). A tutorial on support vector regression. Statistics and Computing, 14(3), 199–222. https://doi.org/10.1023/B:STCO.0000035301.49549.88 [14] Taylor, J. W., & McSharry, P. E. (2007). Short-term load forecasting methods: An evaluation based on European data. IEEE Transactions on Power Systems, 22(4), 2213–2219. https://doi.org/10.1109/TPWRS.2007.907583 [15] Wang, H., Liu, Q., & Guo, Y. (2021). Review on applications of machine learning algorithms in electric load forecasting. Energy AI, 5, 100085. https://doi.org/10.1016/j.egyai.2021.100085 [16] Zhou, K., Yang, S., & Shao, Z. (2020). Review of electric load forecasting models: Techniques and applications. Applied Energy, 285, 116378. https://doi.org/10.1016/j.apenergy.2020.116378

Copyright

Copyright © 2025 Gunjan Jain, S C Choube, Deena Lodwal Yadav. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.