Data Driven Electrical Load Forecasting: A Comapartive Study of Normalization Techniques

Abstract

Short-term electrical load forecasting is of great importance for power system operation, efficient energy management, and planning. For better performance of deep learning models for short-term load forecasting, the quality of the data must be high. In the preprocessing stage, data normalization plays an important role. The present study performs the analysis and comparison of various data normalization techniques, namely Min-Max Scaling, Mean Normalization, Z-Score Standardization, and Gaussian Normalization, for the forecasting of short-term electrical load using an LSTM model. The performance metrics Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R²) are used to compare the performance of the normalization methods. The results show the impact of normalization techniques on the load prediction performance of the LSTM model.

Introduction

This paper focuses on short-term electricity load forecasting using Long Short-Term Memory (LSTM) networks and investigates how different data normalization techniques affect forecasting accuracy.

Background

Accurate electricity demand forecasting is essential for efficient power generation, reducing operational costs, maintaining grid stability, and avoiding power shortages or excess generation. In developing countries, electricity demand continues to grow due to population increase, urbanization, industrialization, and widespread use of electrical appliances. Weather conditions, particularly ambient temperature, significantly influence electricity consumption because higher temperatures increase air-conditioning usage. Electricity demand also exhibits strong daily and weekly seasonal patterns.

Traditional forecasting methods such as linear regression, ARIMA, and exponential smoothing have limitations in capturing complex nonlinear relationships and time dependencies. LSTM networks overcome these limitations by effectively learning long-term temporal patterns, making them well suited for electricity load forecasting.

Methodology

The study uses one year of hourly electricity load and temperature data. Besides historical load values, additional input features include:

• Hour of the day
• Day of the week
• Lagged load values
• Ambient temperature

The main objective is to compare the performance of four normalization methods used with an LSTM model:

1. Min-Max Normalization
2. Mean Normalization
3. Z-Score Normalization
4. Gaussian Function Normalization

Model performance is evaluated using:

• Root Mean Square Error (RMSE)
• Mean Absolute Error (MAE)
• Mean Absolute Percentage Error (MAPE)
• Coefficient of Determination (R²)

Data Preprocessing

Before training, the dataset undergoes preprocessing:

• Missing values are filled using linear interpolation.
• Outliers are detected using the Interquartile Range (IQR) method and replaced or removed.
• Input features are normalized to improve model stability and learning efficiency.

Normalization Techniques

• Min-Max Normalization: Scales values between 0 and 1, improving convergence but sensitive to outliers.
• Mean Normalization: Centers data around zero while preserving relative distances but is also affected by outliers.
• Z-Score Normalization: Standardizes data to a mean of 0 and standard deviation of 1, making it more robust to outliers and improving model generalization.
• Gaussian Function Normalization: Applies a nonlinear Gaussian transformation, suitable for data that approximately follows a normal distribution.

Conclusion

This paper explores the impact of various data normalization techniques on the short-term electrical load forecasting performance of a long short-term memory (LSTM) model. We conducted the evaluation of the four normalization methods (Min-Max Scaling, Mean Normalization, Z-Score Standardization, and Gaussian Normalization) using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R²). The experimental results showed that Gaussian Normalization achieved the lowest RMSE, MAE and MAPE, which indicates lower prediction errors but the lower value of R² indicated that the model was less capable of capturing the variations in the actual load pattern. However, Min-Max, Mean Normalization and Z-Score Standardization gave higher values of R2 which means better predictive performance and consistency. The results show that the choice of normalization method has a large impact on the performance of LSTM- based electrical load forecasting. Furthermore, more advanced deep learning architectures such as Gated Recurrent Unit (GRU), Transformer models, or hybrid LSTM models may enhance a model’s capacity to learn complex temporal dependencies and non-linear relationships in electricity load data.

References

[1] J. He, K. Yuan, Z. Zhong, and Y. Sun, “Enhancing Short-Term Power Load Forecasting With a TimesNet-Crossformer-LSTM Approach,” IEEE Access, vol. 12, pp. 56774–56788, 2024, doi: 10.1109/ACCESS.2024.3383912. [2] T. Hong and S. Fan, “Probabilistic electric load forecasting: A tutorial review,” International Journal of Forecasting, vol. 32, no. 3, pp. 914–938, Jul. 2016, doi: 10.1016/j.ijforecast.2015.11.011. [3] P. Balachandra and V. Chandru, “Modelling electricity demand with representative load curves,” Energy, vol. 24, no. 3, pp. 219–230, Mar. 1999, doi: 10.1016/S0360-5442(98)00096-6. [4] A. Azeem, I. Ismail, S. M. Jameel, and V. R. Harindran, “Electrical Load Forecasting Models for Different Generation Modalities: A Review,” IEEE Access, vol. 9, pp. 142239–142263, 2021, doi: 10.1109/ACCESS.2021.3120731. [5] A. Berrada, K. Loudiyi, and I. Zorkani, “Profitability, risk, and financial modeling of energy storage in residential and large scale applications,” Energy, vol. 119, pp. 94–109, Jan. 2017, doi: 10.1016/j.energy.2016.12.066. [6] S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, Nov. 1997, doi: 10.1162/neco.1997.9.8.1735. [7] K. D. Martinez-Zapata et al., “Data-Driven Load Forecasting in Microgrids: Integrating External Factors for Efficient Control and Decision-Making,” Energies, vol. 19, no. 2, p. 555, Jan. 2026, doi: 10.3390/en19020555. [8] K. Ijaz, Z. Hussain, J. Ahmad, S. F. Ali, M. Adnan, and I. Khosa, “A Novel Temporal Feature Selection Based LSTM Model for Electrical Short-Term Load Forecasting,” IEEE Access, vol. 10, pp. 82596–82613, Jan. 2022, doi: 10.1109/access.2022.3196476. [9] N. Aqilah, S. A. Zaki, A. Hagishima, H. B. Rijal, and F. Yakub, “Analysis on electricity use and indoor thermal environment for typical air-conditioning residential buildings in Malaysia,” Urban Climate, vol. 37, p. 100830, May 2021, doi: 10.1016/j.uclim.2021.100830. [10] H. Dong, Y. Gao, X. Meng, and Y. Fang, “A Multifactorial Short-Term Load Forecasting Model Combined With Periodic and Non-Periodic Features - A Case Study of Qingdao, China,” IEEE Access, vol. 8, pp. 67416–67425, Jan. 2020, doi: 10.1109/access.2020.2986031. [11] L. Lin, L. Xue, Z. Hu, and N. Huang, “Modular Predictor for Day-Ahead Load Forecasting and Feature Selection for Different Hours,” Energies, vol. 11, no. 7, p. 1899, Jul. 2018, doi: 10.3390/en11071899. [12] R. Srinivasan, V. Balasubramanian, and B. Selvaraj, “Short-Term Forecasting of Load and Renewable Energy Using Artifical Neural Network,” International Journal of Engineering Trends and Technology - IJETT, vol. 69, 2021, doi: 10.14445/22315381/IJETT-V69I6P226. [13] T. A. Alghamdi and N. Javaid, “A Survey of Preprocessing Methods Used for Analysis of Big Data Originated From Smart Grids,” IEEE Access, vol. 10, pp. 29149–29171, Jan. 2022, doi: 10.1109/access.2022.3157941. [14] M. Lepot, J.-B. Aubin, and F. H. L. R. Clemens, “Interpolation in Time Series: An Introductive Overview of Existing Methods, Their Performance Criteria and Uncertainty Assessment,” Water, vol. 9, no. 10, p. 796, Oct. 2017, doi: 10.3390/w9100796. [15] V. Chandola, A. Banerjee, and V. Kumar, “Anomaly detection: A survey,” ACM Comput. Surv., vol. 41, no. 3, p. 15:1-15:58, Jul. 2009, doi: 10.1145/1541880.1541882. [16] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. in Adaptive Computation and Machine Learning series. Cambridge, MA, USA: MIT Press, 2016. Accessed: Jul. 10, 2026. [Online]. Available: https://mitpress.mit.edu/9780262035613/deep-learning/ [17] J. M. H. Pinheiro et al., “The Impact of Feature Scaling in Machine Learning: Effects on Regression and Classification Tasks,” IEEE Access, vol. 13, pp. 199903–199931, Jan. 2025, doi: 10.1109/access.2025.3635541. [18] S. Lu and T. Bao, “Short-Term Electricity Load Forecasting Based on NeuralProphet and CNN-LSTM,” IEEE Access, vol. 12, pp. 76870–76879, 2024, doi: 10.1109/ACCESS.2024.3407094. [19] P. Koukaras and C. Tjortjis, “Data Preprocessing and Feature Engineering for Data Mining: Techniques, Tools, and Best Practices,” AI, vol. 6, no. 10, p. 257, Oct. 2025, doi: 10.3390/ai6100257. [20] Niño-Adan, Iratxe & Portillo, Eva & Landa-Torres, Itziar & Manjarres, Diana. (2021). Normalization Influence on ANN-Based Models Performance: A New Proposal for Features’ Contribution Analysis. IEEE Access. 9. 125462-125477. 10.1109/ACCESS.2021.3110647. [21] J. Sola and J. Sevilla, “Importance of input data normalization for the application of neural networks to complex industrial problems,” IEEE Transactions on Nuclear Science, vol. 44, no. 3, pp. 1464–1468, Jun. 1997, doi: 10.1109/23.589532. [22] Sujon, Khaled & Hassan, Rohayanti & Towshi, Zeba & Othman, Manal & Samad, Md Abdus & Choi, Kwonhue. (2024). When to Use Standardization and Normalization: Empirical Evidence From Machine Learning Models and XAI. IEEE Access. 12. 135300-135314. 10.1109/ACCESS.2024.3462434. [23] N. Passalis, A. Tefas, J. Kanniainen, M. Gabbouj, and A. Iosifidis, “Deep Adaptive Input Normalization for Time Series Forecasting,” IEEE Trans Neural Netw Learn Syst, vol. 31, no. 9, pp. 3760–3765, Sep. 2020, doi: 10.1109/TNNLS.2019.2944933. [24] S. K.n., L. Pinto, S. Gopalan, and P. Balasubramaniam, “Equivalence class and modified Gaussian methods for normalization of time series data on AI models,” Expert Systems with Applications, vol. 277, p. 127166, Jun. 2025, doi: 10.1016/j.eswa.2025.127166. [25] Rousseeuw, Peter & Hubert, Mia. (2018). Anomaly Detection by Robust Statistics. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery. 8. 10.1002/widm.1236. [26] Tunnicliffe Wilson, Granville. (2016). Time Series Analysis: Forecasting and Control, 5th Edition, by George E. P. Box, Gwilym M. Jenkins, Gregory C. Reinsel and Greta M. Ljung, 2015. Published by John Wiley and Sons Inc., Hoboken, New Jersey, pp. 712. ISBN: 978?1?118?67502?1. Journal of Time Series Analysis. 37. 709-711. 10.1111/jtsa.12194. [27] N. Passalis, A. Tefas, J. Kanniainen, M. Gabbouj, and A. Iosifidis, “Deep Adaptive Input Normalization for Time Series Forecasting,” Sep. 22, 2019, arXiv: arXiv:1902.07892. doi: 10.48550/arXiv.1902.07892. [28] “Investigating the Impact of Data Normalization Methods on Predicting Electricity Consumption in a Building Using different Artificial Neural Network Models | Request PDF,” ResearchGate, doi: 10.1016/j.scs.2024.105570. [29] Gers, Felix & Schmidhuber, Jürgen & Cummins, Fred. (2000). Learning to Forget: Continual Prediction with LSTM. Neural Computation. 12. 2451-2471. 10.1162/089976600300015015. [30] Y. Wang, N. Zhang, and X. Chen, “A Short-Term Residential Load Forecasting Model Based on LSTM Recurrent Neural Network Considering Weather Features,” Energies, vol. 14, no. 10, p. 2737, Jan. 2021, doi: 10.3390/en14102737. [31] R. J. Hyndman and A. B. Koehler, “Another look at measures of forecast accuracy,” International Journal of Forecasting, vol. 22, no. 4, pp. 679–688, Oct. 2006, doi: 10.1016/j.ijforecast.2006.03.001. [32] T. Hong, P. Pinson, and S. Fan, “Global Energy Forecasting Competition 2012,” International Journal of Forecasting, vol. 30, no. 2, pp. 357–363, Apr. 2014, doi: 10.1016/j.ijforecast.2013.07.001. [33] L. G. Rocha, S. Gomes Soares Alcala, and L. P. Garces Negrete, “Short-term electric load forecasting using neural networks: A comparative study,” in 2020 IEEE PES Transmission & Distribution Conference and Exhibition - Latin America (T&D LA), Montevideo, Uruguay: IEEE, Sep. 2020, pp. 1–6. doi: 10.1109/TDLA47668.2020.9326196. [34] A. D. Myttenaere, B. Golden, B. L. Grand, and F. Rossi, “Mean Absolute Percentage Error for regression models,” Neurocomputing, vol. 192, pp. 38–48, Jun. 2016, doi: 10.1016/j.neucom.2015.12.114. [35] Chicco, Davide & Warrens, Matthijs & Jurman, Giuseppe. (2021). The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation. PeerJ Computer Science. 7. e623. 10.7717/peerj-cs.623.

Copyright

Copyright © 2026 Kuldeep Atmdarshi, Seema Pal. 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.