Authors: Chawalit Jeenanunta; K. Darshana Abeyrathna
Addresses: School of Management Technology, Sirindhorn International Institute of Technology, Thammasat University, Pathum Thani, Thailand ' School of Management Technology, Sirindhorn International Institute of Technology, Thammasat University, Pathum Thani, Thailand
Abstract: Short-term load forecasting is to forecast the next day electricity demand for 48 periods and it is important to make decisions related to the electricity generation and distribution. Neural network (NN) is selected for forecasting the future electricity consumption since its ability of recognising and learning nonlinear patterns of data. This research proposes the combination usage of genetic algorithm (GA) to train the neural network and results are compared with the results from backpropagation. Data from the electricity generating authority of Thailand (EGAT) is used in this research to demonstrate the performance of the proposed technique. The dataset contains weekday (excluding Mondays) load demand from 1st of October to 30th of November 2013. November load is forecasted using an NN with 192 inputs and 48 outputs. Even though GA takes more time for training neural networks, it gives better results compared to backpropagation.
Keywords: genetic algorithm; GA; encoding; neural network; NN; optimising neural network's weights; forecasting; short-term load forecasting; STLF.
International Journal of Energy Technology and Policy, 2019 Vol.15 No.2/3, pp.337 - 350
Received: 06 Mar 2017
Accepted: 05 Jul 2017
Published online: 04 Mar 2019 *