To populate Scholarly, sign in here .


Title Performance Analysis of Artificial Neural Networks Training Algorithms and Activation Functions in Day-Ahead Base, Intermediate, and Peak Load Forecasting
Posted by Lemuel Clark Velasco
Authors Velasco, Lemuel Clark ; Estoperez, Noel ; Jayson, Renbert Jay; Sabijon, Caezar Johnlery; Sayles, Verlyn
Publication date 2019
Conference Future of Information and Communication Conference
Volume 70
Pages 284-298
Publisher Springer, Cham
Abstract Artificial Neural Networks (ANN) has been highly utilized in short term electric load forecasting not just among aggregated consumed load but also in predicting the specified base, intermediate and peak loads. To ensure success in its predictive capability, every ANN model implementation should employ the appropriate training algorithm and activation function that will be suitable to the historical data that it is processing. This study conducted performance analysis of six models having different combination of training algorithms namely Quick Propagation, Resilient Algorithm and Back Propagation and activation functions namely Gaussian and Sigmoid. Electric load data preparation was conducted through data correction, Min-Max data normalization and clustering to identify the base, intermediate and peak loads. After determining the ANN models’ input, hidden and output neurons from its respective layers, the ANN model having the combination of Quick Propagation training algorithm and Gaussian activation function yielded the lowest MSE and MAPE values having 0.005700397 and 5.88% respectively. The day-ahead base, intermediate, and peak load forecasting model developed in this study has the potential to be implemented in order to suffice the need of electric power systems in predicting the necessary system loads for their economic decisions, power dispatching, system planning, and reliability evaluation.
Index terms / Keywords Artificial neural network Electric load forecasting Base intermediate and peak load