Hour-Ahead Electric Load Forecasting Using Artificial Neural Networks

Autor: Justine Shane S. Macarat, Lemuel Clark P. Velasco, Karl Anthony S. Arnejo, Mia Amor C. Tinam-isan
Rok vydání: 2021
Předmět:
Zdroj: Proceedings of Sixth International Congress on Information and Communication Technology ISBN: 9789811617805
ICICT (3)
DOI: 10.1007/978-981-16-1781-2_73
Popis: To ensure an efficient supply of electricity, power utility companies make use of a combination of either short-term, medium-term, and long-term forecasting techniques. This paper presents a strategy for implementing hour-ahead electricity load forecasting using artificial neural networks (ANN). Data preparation through data selection, cleaning, partitioning, and transformation was performed to the dataset provided by a power utility in Southern Philippines. ANN models using a feedforward architecture having 9 input neurons, 6 hidden neurons, and 1 output neuron with a learning rate set to 0.00001, momentum set to 0.7 with an epoch value of 20,000, and a maximum error of 0.001 was implemented using a Java-based open-source library. Results showed that the ANN model with a quick propagation training algorithm and sigmoid activation function had a Mean Absolute Percentage Error (MAPE) of 2.91%, and the ANN model with resilient propagation training algorithm and sigmoid activation function had a MAPE of 3.49%. This study shows that with appropriate data preparation and machine learning implementation, ANN has the potential to aid the decision-making of power utility companies through short-term forecasting in terms of hour-ahead electric load consumption prediction.
Databáze: OpenAIRE