Forecasting of the Daily Peak Electrical Load of a Machine-Building Enterprise

Abstract

The factors of the daily peak electrical load of a machine-building enterprise are investigated and an approach to its forecasting is proposed. Data of the Minsk Automobile Plant for 2024–2025 (731 days), together with meteorological and monthly production indicators, were used. The calculations were performed in the authors' adaptive system for comprehensive statistical analysis and forecasting (ASCSAF). Seven production-calendar mismatches and 12 violations of the relationship between electricity consumption and peak load were detected and eliminated; without their elimination the forecast error increases by about 25 %. Factor analysis identified three factors explaining 62.6 % of the variance. A combined scheme was adopted: ridge regression for working days, random forest for non-working days, and adaptive correction. The MAPE on working days was 4.75 % one day ahead and 7.14 % one month ahead.