<![CDATA[The consumption of electrical power continues to increase every day, along with technological advances, the increase in the use of machines that depend on electricity and the growth of the human population. Prediction of electric power consumption is difficult due to various things such as weather conditions and dynamic behavior of residents. So to overcome this, deep learning methods are used, namely CNN (convolutional neural network), LSTM (long-short term memory), BDLSTM (bi-directional long short term memory), CNN-LSTM and CNN-M-BDLSTM with three stages. The first stage is fixing the missing value. The second stage performs data processing and building model from deep learning methods. The third stage evaluates the prediction results with error metrics. From the results of the analysis, the CNN method produces the smallest loss and training time of 0.0010 9/step ms, the LSTM and CNN-LSTM methods produce the largest loss of 0.0019 while the longest training time is 139/step ms in the BDLSTM method. Method The largest MSE, MAE, and MAPE values are 0.150, 0.258 and 0.176 respectively the results of the CNN method. The smallest MSE and MAE values are 0.082 and 0.174 the results of the BDLSTM method. While the smallest MAPE value obtained by the CNN-M-BDLSTM method is 0.148. This proves that the ability to predict short-term electric power consumption using the deep learning method has a good ability in terms of the error metric results.]]>
Copyrights © 2024
