Radial basis function (RBF) neural network for load forecasting during holiday

Providing solution for short term load forecasting is a major challenge remained for researchers due to the nature characteristics of load which are non-linear, probabilistic and uncertainty. As the statistical assumption may fail to estimate the load profile precisely, the intelligent techniques play important role to provide alternative solutions. This paper discusses the variant of artificial neural network called radial basis function (RBF) neural network for short term load forecasting. The method is recently attracted attention due to structure simplicity and high identification performance. The RBF method is an artificial neural network model motivated by locally-tuned response biological neurons that provide selective response characteristics for some finite range of the input signal space. The estimation process is carried out with 4 previous peak load holiday to predict the peak load of the next holiday using data of the year 2005–2011 in Makassar City, Indonesia. The validation results show that the proposed method can offer very accurate forecasting results, indicated by small mean absolute percentage error (MAPE) for the estimation task of the year of 2012 and 2013 in comparison to conventional least square polynomial approximation method.