Implementation of Encoder-Decoder based Long Short-Term Memory Network for Short-Term Electrical Load Forecasting

Electrical load forecasting has extensive importance in power system designing and optimization. Keeping pace with the growing economy and moving towards a digitalized network, Bangladesh Power System (BPS) has to be renovated in terms of efficient load forecasting strategies. To meet the issue, this research proposes an effective short-term load forecasting (STLF) technique based on a unique form of recurrent neural network (RNN) known as the Long Short-Term Memory (LSTM) network. A new dataset gathered from a public owned power company of Bangladesh, is applied into the model which will help to get a better and recent understanding of BPS load. Employing this viable method, one hour ahead electrical load of BPS can be forecasted easily with a minimized error rate (MAPE 2.29%). In addition, Random Forest (RF) regression model is used to compare with the forecasting results of proposed LSTM. Again, the proposed LSTM model is used to get predictions for the different months of a given year to visualize how much load fluctuation of BPS load takes place in different seasons. In case of STLF, the prediction outcomes on the Bangladesh’s electricity load demand, indicate that the proposed model can work flawlessly in grid optimization with greater precision.