An Intelligent ANN Approach for Short Term Electric Load Forecasting

Medha Joshi, Rajiv Singh

Abstract: As conventional approaches like the regression model and the time-series based models, are not much suitable because of the complexity and labour involved in modelling. So in this proposed work, to fulfil the requirement of accurate electric load forecast, an Artificial Neural Network (ANN) approach was used to forecast the next hour electric load for Safdarjang, New Delhi region. To accomplish the abovementioned task, initially appropriate weather variables were selected for the aforementioned region, on the basis of correlation coefficient. Then different forecasting models with several combinations of weather variables were prepared and tested. Simplest and well known ANN i.e. Multilayer Feed-Forward Neural Network with back-propagation algorithm was used. Finally, from this research work, appropriate input variables were specified to forecast the next hour electric load. The forecasting model, containing temperature and pressure as input weather variables was most appropriate one for the aforementioned region. An important observation of this research work was that those weather variables were appropriate for accurate forecast, which were highly correlated with load.

Keywords: Artificial neural network, back-propagation algorithm, input variables, multilayer feed-forward neural network, short term electric load forecasting