Multi-substation control central load area forecasting by using HP-filter and double neural networks (HP-DNNs)

Electricity load demand forecasting of Thailand using Hodrick-Prescott (HP) filters and double-neural networks (DNNs) is presented in this article by dividing whole country area into multi-substation areas. The signals of load demand in each subarea will be decomposed to trend and cycling signals by HP-filter before sent to DNNs for load demand forecast. The trend signals show close relationship with economic affecting features, while the cycling signals demonstrate strong relationship with weather features. These obvious correlations will be used for feature input selections. In the finally stage, the forecasting results from each subarea will be composed for the whole country area result. Comparing to other forecasting models, this approach not only reduce complexity of the forecasting model but also decrease mean absolute percent error (MAPE) as 1.42%. Moreover, this method can be applied to other load forecasting in power system and any application that can be separated into subarea. (C) 2012 Elsevier Ltd. All rights reserved.