Scholarlink

The present article performs short-term electricity demand prediction using deep learning tools for the Ecuadorian electrical system, specifically in the province of Santa Elena, Ecuador. The prediction was carried out using two neural network models: Long Short-Term Memory (LSTM) and Fully Connected Neural Network (FCNN). Additionally, exogenous variables such as temperature and relative humidity were incorporated. Historical data of exogenous variables were obtained from the NASA POWER | DAVe website with the aim of exploring the relationship between these variables and energy consumption. Subsequently, data processing was conducted, and both neural networks were trained. The results indicate robustness in electricity demand prediction for both architectures; however, the LSTM network exhibits significantly higher accuracy, evidenced by a Coefficient of Determination ( \({\text{R}}^{2}\) ) of 0.986 compared to FCNN. These findings suggest that LSTM has an enhanced capacity to model and anticipate electricity demand, as it considers the correlation with exogenous variables that affect electricity consumption.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Short-Term Prediction of Electricity Demand Using Deep Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Óscar Gómez,
  • Manuel Montaño,
  • Washington Torres,
  • José Carguachi,
  • Dario Gutierrez,
  • German Castellanos

摘要

The present article performs short-term electricity demand prediction using deep learning tools for the Ecuadorian electrical system, specifically in the province of Santa Elena, Ecuador. The prediction was carried out using two neural network models: Long Short-Term Memory (LSTM) and Fully Connected Neural Network (FCNN). Additionally, exogenous variables such as temperature and relative humidity were incorporated. Historical data of exogenous variables were obtained from the NASA POWER | DAVe website with the aim of exploring the relationship between these variables and energy consumption. Subsequently, data processing was conducted, and both neural networks were trained. The results indicate robustness in electricity demand prediction for both architectures; however, the LSTM network exhibits significantly higher accuracy, evidenced by a Coefficient of Determination ( \({\text{R}}^{2}\) ) of 0.986 compared to FCNN. These findings suggest that LSTM has an enhanced capacity to model and anticipate electricity demand, as it considers the correlation with exogenous variables that affect electricity consumption.