Decomposition and statistical analysis for regional electricity demand forecasting

This paper proposes a decomposition approach for modelling the electricity demand trend and variability for medium- and long-term forecasting. This approach decomposes the historical time series into a number of components according to seasonality and day of week. For each component, the yearly and intra-season trends are identified by regression analysis, and the diurnal demand pattern and its associated variability are determined by statistical estimates. Because the decomposition is in line with the changes in seasonality, day of week, and daily activity, the demand models as derived conform to the intuitive interpretation for temporal changes of demand levels. In contrast to most existing methods, this approach does not require involved structural models or time series analysis, saving the efforts of complex non-linear parameter estimations, and is relatively easy for implementation. We apply the proposed approach to half-hourly electricity demand data recorded from 2002 to 2011 for the states of Queensland, Victoria, and the South East Queensland region, Australia. We compare the results for South East Queensland from Monte Carlo simulation with the historical demand, and use it for annual average and peak electricity demand projection up to 2020.

► A decomposition approach for electricity demand trend and variability modelling.
►  The decomposition depends on seasonality and day of week.
►  The demand variability is modelled by normal distributions.
►  Implementation of the derived demand models is straightforward for simulation.
►  Suitable for mid- and long-term probabilistic average and peak demand forecasting.