Models for long-term energy forecasting

Based on historical data related to an actual power system, the paper develops and evaluates the accuracy of a range of mathematical models for long-term forecasting of energy demand in the system. Starting from the basic relationship in an econometric model based on regression analysis, the development and evaluation are extended to include advanced and recently-proposed methods which use dynamical functional-link net (FLN) and wavelet networks. The constrained optimisation technique based on sequential quadratic programming (SQP) is applied to identify the parameters of the forecasting models. Extensive testing indicates that the model using wavelet functions gives the best performance in terms of forecasting accuracy. Finally, the effect of temperature on energy demand is incorporated in this model, which leads to a probabilistic method of long-term energy demand forecasting.