Electricity Price Forecasting One Day Ahead by Employing Hybrid Deep Learning Model
Abdallah Abdellatif, Hamza Mubarak, Shameem Ahmad, Saad Mekhilef, Hamdan Abdellatef, Hazlie Mokhlis, Jeevan Kanesan
Abstract
This study proposes hybrid Deep Learning (DL) models for electricity price forecasting (EPF) one day ahead of the Nord Pool spot electricity market. The proposed hybrid DL model employs the Bidirectional Long Short-Term Memory (BiLSTM) and Convolution Neural Network (CNN) to identify short-term local dependence trends among parameters and to discover long-term correlations for time series trends. In addition, a linear bypass (auto-regressive) is included to address the neural network model's scale insensitivity issue. The proposed model (CNN-BiLSTM-AR) is compared with different models like CNN-LSTM, CNN-BiLSTM, and CNN-LSTM-AR, which were evaluated by employing different performance indicators including Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE). The findings revealed that the suggested CNN-BiLSTM-AR model achieved the lowest RMSE and MAE with values of 8.67 and 5.43, respectively, followed by CNN-BiLSTM and CNN-LSTM-AR with a value of 9.08, 9.154 and 5.65, 5.645 for RMSE and MAE. Further, the CNN-LSTM model performed badly, attaining the highest RMSE and MAE values of 9.471 and 5.899, respectively.