Scholarlink

This study presents a hybrid deep learning model for short- and long-term wind power forecasting based on the integration of Long Short-Term Memory (LSTM) networks and Extreme Gradient Boosting (XGBoost). The dataset, collected from the official meteorological monitoring network of Tungurahua, Ecuador, covers hourly records from January 2024 to December 2025, including wind speed, temperature, humidity, and solar radiation. After preprocessing, time-lagged features and normalization were applied to enhance the temporal coherence of the data. The standalone LSTM model achieved a Root Mean Square Error (RMSE) of 0.1706 and a coefficient of determination (R2) of 0.4388 during external validation. In contrast, the hybrid LSTM + XGBoost configuration improved the prediction accuracy, obtaining an RMSE of 0.1560 and an R2 of 0.5306. The annual projection for 2026 shows wind power fluctuations between 30 kW and 200 kW, consistent with real wind dynamics in the Andean region. The results demonstrate that the hybrid model effectively captures nonlinear and temporal dependencies, providing a stable and interpretable framework for renewable energy management.

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

Wind Energy Prediction Based on Machine Learning Methods

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Juan Acosta,
  • Alex Collaguazo,
  • Jessica N. Castillo,
  • E. Freddy Robalino,
  • Luigi O. Freire,
  • Luis Antonio Flores

摘要

This study presents a hybrid deep learning model for short- and long-term wind power forecasting based on the integration of Long Short-Term Memory (LSTM) networks and Extreme Gradient Boosting (XGBoost). The dataset, collected from the official meteorological monitoring network of Tungurahua, Ecuador, covers hourly records from January 2024 to December 2025, including wind speed, temperature, humidity, and solar radiation. After preprocessing, time-lagged features and normalization were applied to enhance the temporal coherence of the data. The standalone LSTM model achieved a Root Mean Square Error (RMSE) of 0.1706 and a coefficient of determination (R2) of 0.4388 during external validation. In contrast, the hybrid LSTM + XGBoost configuration improved the prediction accuracy, obtaining an RMSE of 0.1560 and an R2 of 0.5306. The annual projection for 2026 shows wind power fluctuations between 30 kW and 200 kW, consistent with real wind dynamics in the Andean region. The results demonstrate that the hybrid model effectively captures nonlinear and temporal dependencies, providing a stable and interpretable framework for renewable energy management.