A novel hybrid forecasting scheme for electricity demand time series

Electricity demand/load forecasting always plays a vital role in the management and operation of power systems, since it can help develop an optimal action program for power producers, end-consumers and government entities. Inaccurate prediction may cause an additional production or waste of resources due to high operational costs. This paper investigated the benefit of combining data features to produce short-term electricity demand forecast. The nature of the electricity usually presents the complex characteristic and obvious seasonal tendency. In this paper, the advantage of adaptive Fourier decomposition is firstly used to extract the fluctuation characteristics. Then, the condition of the linear and stationary sequence is satisfied and the sub-series are performed to measure and eliminate the seasonal pattern. In the process of seasonal adjustment, the average periodicity length is identified quantitatively. In addition, to realize the generalization performance on real electricity demand data, the sine cosine optimization algorithm is applied to select the penalty and kernel parameters of support vector machine. The empirical study showed that the superior property of the proposed hybrid method profits from the effect of data pretreatment and the findings prove that this hybrid modeling scheme can yield promising prediction results within acceptable computational complexity.