Application of artificial neural network for short term electricity demand forecasting

Abstract

Many researchers around the world work on short term electricity demand forecasting (STLF) in order to establish an accurate power planning and generation system in their countries. This research, with its focus on short-term load forecasting, aims to fill this gap by implementing two methodologies based on Artificial Neural Network (ANN) and Autoregressive Integrated Moving Average (ARIMA) applied on a set of half an hourly load demand data of six years, provided by Ceylon Electricity Board in Sri Lanka. The data of first five years (~70% of the dataset) were used to train the algorithms and those of the last year (~30% of the dataset) were used for testing. The effect of historical load demand patterns on making the prediction of the next 24 hours were studied. Moreover, with the historical data, unlike in most literature which forecasts only one value (either peak load demand of the day or only the load demand of the next half an hour), the demand of the entire day (48 values for each half an hour) is forecasted at once. The predictions obtained by the application of ANN were compared with those of ARIMA methodology which is a benchmark of comparing predictions in STLF. None of the applications provided deviated predictions compared to each other and ANN can be used to predict the next day half-hourly electricity demand since the application was successful in grasping the periodic patterns that exist in half hourly series.