Forecasting electricity consumption: a comparison of growth curves, econometric and ARIMA models for selected countries and world regions.
Thesis DisciplineElectrical Engineering
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
This thesis presents six forecasting models for annual electricity consumption based on various time series extrapolation techniques. The proposed models are based on growth curves, multiple linear regression analysis using economic and demographic variables (referred to as the Combined model) and autoregressive integrated moving average (ARIMA) techniques. The proposed models are applied to electricity consumption data of New Zealand, the Maldives, the United States of America and the United Kingdom. The models are also applied to the electricity consumption data of various world regions and the world total, and are compared using model fit and forecasting accuracies. This thesis initially investigates the patterns of electricity consumption to study the link between electricity consumption, economic growth and population. Although the link between economic growth and electricity consumption varies between developing and industrialised countries, the link is strong enough to justify the use of these variables in the models of all countries and regions. In addition, the patterns appear uninfluenced by the adoption of regulatory or market type economies, suggesting that the forecasts of the proposed models should not be affected during the period of regulatory reforms in the electricity industry. In general, application of the models at the country level revealed that the simple Harvey model, based on a growth curve, has performed better than the more complex ARIMA and regression models. For the regional and world total electricity consumptions, the ARIMA models are the best followed very closely by the regression and Harvey models. However, Harvey is the only model that gave among the best forecasts in the short, medium and long term forecasting. Overall, it was concluded that the simple Harvey model performed better than or as good as the more complex ARIMA and Combined models. In general, the Harvey model is the best in forecasting mature electricity industries when more data points are available, the ARIMA model is the best when the number of data points available is limited and the Combined model always gave average results for all data sets.