ORCHESTRATING ELECTRICITY CONSUMPTION PREDICTIONS WITH SARIMA, DYNAMIC REGRESSION, AND MINT

Long-term forecasting of electricity consumption is essential for planning in the power sector. This paper proposes a methodology that combines univariate and multivariate models, enabling monthly updates even when macroeconomic scenarios are revised annually. The approach uses univariate models for the short term and regression models based on GDP and sectoral value added for the long term. Applied to a distribution company, the methodology includes forecast reconciliation by consumption class and total using the MinT method, ensuring coherence.