Evaluation of the influence of the bus-bar voltage profiles on demand forecasting by using neural networks

• Autor(a) principal: Fernandes, Amanda Thais dos Reis
• Data de Publicação: 2021
• Outros Autores: Fonseca, Jean Lucas Tourinho, Silva, Iuri Leno Pereira da, Agamez Arias, Piedy Del Mar, Ramos, Rodrigo Andrade, Oliveira, Werbeston Douglas de
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Research, Society and Development
• Texto Completo: https://rsdjournal.org/index.php/rsd/article/view/20917
• topic: Pronóstico de demanda; Muy cirto plazo redes neuronales; Histórico de carga; Perfiles de tensión.; Load forecasting; Very short-term; Neural networks; Load historical data; Voltage profile.; Previsão de carga; Redes neurais; Tensão nas barras; Histórico de carga; Curtíssimo prazo.

The very short-term load forecasting allows operation engineers an economic and safe dispatch of the power system while dynamically contributes to the prices in the energy market. Several methodologies such as regression analysis, time series, machine learning approaches, deep learning methods, and artificial intelligence have been used to forecast load. However, several external factors become the forecasting a more complex task than it initially appears to be. For this reason, neural networks have been presented as a computational intelligence technique capable of dealing with the load forecasting problem with great precision. In this context, this work aims to evaluate the impact of voltage profiles of the power system bus on the load forecasting. For this, it was proposed to study three database arrangements ((1) normalized load historical data; (2) normalized load historical data and voltage profile in load bars; and, (3) normalized load historical data, voltage profile in load bus and seasonality of these loads) to train nine neural networks of the MLP type with two layers. The proposed approach is evaluated based on data obtained from the state estimator of a network of a large company in the northern region of Brazil. The results show that, according to the MSE and MAPE values obtained, all the neural networks evaluated achieve a forecasting of the load as expected. However, the best performance was achieved with the arrangement that considered a database that records a normalized load historical data and voltage profile in load bus.