| Title |
Week-Ahead Load Forecasting using Multilayer Perceptron Neural Network for a Power Utility |
|
Authors |
Velasco, Lemuel Clark; Bokingkito, Paul; Vistal, Jogie; |
|
Publication date |
2017 |
|
Conference |
17th Conference of the Science Council of Asia |
| Pages |
6 |
| Publisher |
National Research Council of the Philippines |
| Abstract |
Week-ahead load prediction which involves seven days in daily resolution of forecasted load is an important function in capacity planning and maintenance scheduling of power utility companies. This study attempted to develop a prediction model which generates week-ahead load using multilayer perceptron neural network (MLPNN), an artificial neural network architecture that models the human brain. Historical load consumption from a local power utility in the Philippines was carefully validated and normalized using min-max normalization technique. Resilient propagation training algorithm was used and compared from the performance of Tanh, Sigmoid and Gaussian activation functions in order to determine which techniques yields a more accurate model to train the network. The results showed that using MLPNN, resilient propagation as learning algorithm and its appropriate activation function could yield to an accurate prediction with minimum error in predicting the week-ahead consumption of the power utility company. From the three activation functions being compared, sigmoid activation function in resilient propagation produces the most efficient and least network error on its training. These findings suggest that a MLPNN architecture with resilient propagation and sigmoid activation function is can be an efficient predictive model in forecasting week-ahead load consumption for a local power utility company. |
| URL |
|
Mindanao State
University - Iligan Institute of Technology