International Journal of Value Engineering

Analysis of factors and forecasting of the daily maximum electric load of an automobile factory

Vishniakou, U.A.

Belarusian State University of Informatics and radioelectronics

Paloska K.I.

Belarusian State University of Informatics and radioelectronics

Keywords: daily maximum load; autuvobile factory; factor analysis; principal component analysis; ridge regression; power consumption forecasting; declared capacity; comprehensive statistical analysis

---

Abstract

The study examines the factors influencing the daily maximum electrical load of an automobile plant and develops an approach to forecasting this indicator over daily, monthly, and nine-month horizons. The study is based on operational data from the Minsk Automobile Plant for 2024–2025, which includes 731 daily observations, meteorological data, and monthly production indicators, including data from the foundry production. The calculations were carried out using the author's comprehensive statistical analysis (ACSA) system, developed in Python. The system provides data loading, verification, and preliminary processing, as well as correlation and factor analysis, training of predictive models, and adaptive adjustment of the forecast. The initial data revealed seven days with discrepancies in the production schedule and twelve days with a violation of the permissible ratio between electricity consumption and the maximum load. Factor analysis of the daily data (n=485, KMO=0.653, p<0.001) made it possible to identify three factors that explain 62.6% of the variance. Eleven configurations of predictive models, including random forest and ridge regression, have been compared. A scheme with division by day type has been adopted: Ridge regression is used for working days, random forest for non-working days, and adaptive adjustment is applied to both. This combination was chosen due to the statistically indistinguishable accuracy of competing models, as it is more interpretable. The mean absolute percentage error for working days was 4.75% for a one-day forecast and 7.14% for a monthly forecast. Organizational measures to smooth out load peaks have been proposed, and their potential effect has been assessed.

---

References

[1] Hong T., Fan S. Probabilistic electric load forecasting: A tutorial review. International Journal of Forecasting. – 2016. – Vol. 32, N 3: 914–938.

[2] Nadtoka, I. I., Sedov A. V. Forecasting electricity consumption of industrial enterprises using neural networks. Proceedings of Higher Educational Institutions. Electromechanics. – 2018. – No. 4: 78–85.

[3] Vishniakou, U. A., Paloska K. I. Software tools for automating calculations for factorial, regression, and correlation analyses. System Analysis and Applied Informatics. – 2026. – No 1: 82–87.

[4] Vishniakou, U.A., Khatskevich G.A., Paloska K.I. Comprehensive analysis for assessing the use of Internet of Things networks in the educational process at the University of Informatics and Radioelectronics. Statistics and Economics. – 2025. – Vol. 22, No. 4: 52–60.

[5] Kaiser, H. F., Kennard R. W. An index of factorial simplicity. Psychometrika. – 1974. – Vol. 39, No 1: 31–36.

[6] Hoerl, A. E. Ridge regression: biased estimation for nonorthogonal problems / A. E. Hoerl, // Technometrics. – 1970. – Vol. 12, No 1: 55–67.

[7] Breiman, L. Random forests. Machine Learning. – 2001. – Vol. 45, No 1: 5–32.

[8] Diebold, F. X., Mariano R. S. Comparing predictive accuracy. Journal of Business & Economic Statistics. – 1995. – Vol. 13, No 3: 253–263.