Effect of neural network structure for daily electricity load forecasting

Abstract

Accurate electricity demand forecasts are critical for daily operations planning. They influence many decisions, including commits to produce electricity for a given period. This paper presents a short term electricity demand forecasting system using the Artificial Neural Networks (ANNs). The model is trained and tested on 30-minutes historical load data with temperature from the Electricity Generating Authority of Thailand (EGAT) from January 1, 2012 to December 31, 2013. The ANNs use historical load data with temperature to forecast daily electricity demand in Thailand. Holidays, bridging holidays, and outliers of the raw data are detected and replaced. Historical load (previous day, previous week), forecasted day total load, forecasted day temperature, previous day temperature, calendar days (Day of week and Month), and whether the forecasted day is a holiday or not are used as input parameters. The forecasting performances are compared with Regression model. Best performance has shown with ANN.