Multivariate Statistical Analysis based Methodology for Long-Term Demand Forecasting

Abstract

Forecasting models are necessaries in electrical
utilities to set the energy cover for several years in order to
minimize the operational cost. In this sense, a low error is required
to avoid high levels energy exchange with market and hence an
increase in the operation costs. However, demand energy has
monthly a behavior correlated with some economics variables
(e.g., gross domestic product, gold price), type and amount of
days, population, historical demand, inter alia. For this reason, it
is necessary to design a statistical methodology that allows
selecting suitable variables and characterize them to establish
which factors are significant for understanding. This paper
proposes a methodology to identify those significant factors in the
monthly forecasting model for energy demand in the power
purchases area. The climatic scenario to increase the sensibility
of the model, when the probability of the “Niño” or “Niña”
phenomena increase month to month, was proposed.