Application of Elman and NCA to uncover most informative parameter for load forecasting

Energy management is one of the critical issues for planning various energy sources for optimum operation. On the other hand, both consumption and production sides must be taken into account while managing. On the consumption side, accurate load forecasting is of vital importance. In this study, the hourly consumption at the main campus area of Afyon Kocatepe University is measured and recorded. It is intended to build a forecasting model for the consumptions. For this aim, the hour, day, month and season of the consumption times besides the past-recorded samples of the consumption are taken into account as the features and they are considered as information carriers about the variation of the load demand. However, it is also aimed in this study to make the most informative parameter noticeable for the forecasting hierarchy. Therefore as a first step, neighborhood component analysis (NCA) is used for determining the most informative feature. Then, Elman Neural Networks (Elman) is used for forecasting of hourly demand. The experimental results show that, the feature selection provides determination of more meaningful features and the proposed approach is successful in load forecasting.