| Title |
Performance Analysis of Different Combination of Training Algorithms and Activation Functions in Predicting the Next Day Electric Load |
|
Authors |
Velasco, Lemuel Clark; Villezas, Christelle; Palahang, Prinz Nikko; Dagaang, Jerald Aldin |
|
Publication date |
2016 |
|
Conference |
16th Philippine Computing Science Congress |
| Pages |
6 |
| Publisher |
Computing Society of the Philippines |
| Abstract |
Artificial Neural Network (ANN) is a widely used pattern recognition technique in predicting the next day electric load of power utility companies. ANN has the ability to predict the next day electric load if the appropriate training algorithm and activation function is used. In this study, using Fast Artificial Neural Network (FANN) library, nine models having different combinations of training algorithms namely Batch, Incremental and RPROP and activation functions namely Elliot, Gaussian and Sigmoid were compared. This study compared data normalization techniques and showed that Min-Max normalization yielded smaller Mean Absolute Percentage Error (MAPE) values compared to Max normalization. Moreover, out of the nine models tested using Root Means Squared Error (RMSE) and MAPE, Model 2 which used Incremental training algorithm and Sigmoid activation function yielded the lowest MAPE and RMSE. The results obtained in this research clearly suggest that ANN model is a viable forecasting technique for a next day electric load forecasting system. |
| URL |
|
Mindanao State
University - Iligan Institute of Technology