Scholarlink

<p>The growing integration of renewable energy into modern grids demands accurate wind power forecasting to ensure stable operations and efficient energy management. Existing methods often overlook nonlinear patterns and fail to fully exploit the Internet of Things (IoT)-based data acquisition, along with advanced deep learning-driven optimization for grid-connected power prediction and optimization. This research proposes a novel framework for grid-connected wind power prediction and optimization based on the IoT and deep learning. IoT-enabled sensors collect real-time meteorological and turbine operational data, which undergo enhanced preprocessing using the min-max normalization. Key predictive features are extracted using the Kernel Principal Component Analysis (Kernel-PCA) to capture nonlinear feature relationships and reduce dimensionality. The proposed Fennec Fox optimizer-tuned Dynamic Bidirectional Long Short-Term Memory with Attention Mechanism (FF-DBLSTM-AM) is employed to perform accurate grid-connected wind power prediction and optimization. The Fennec Fox optimizer ensures effective hyperparameter tuning, while the DBLSTM-AM architecture captures bidirectional temporal dependencies and applies attention to critical features for enhanced forecasting performance. The outputs are integrated into a grid connected optimization layer employing dynamic load balancing and energy dispatch strategies, enabling improved scheduling and reduced grid imbalance. Implemented in Python, the FF-DBLSTM-AM approach outperforms existing methods such as CNN-BIGRU, achieving an R-squared value of 0.9465(94.65%), RMSE of 0.1016, MAE of 0.0401, and MSE of 0.0112, demonstrating higher predictive accuracy and improved grid stability. These findings provide a reliable and scalable solution fo<sup>r</sup> enhancing wind energy integration on smart grids and highlight the advantages of the proposed model over conventional deep learning architectures.</p>

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 power grid-connected power prediction and optimization based on Internet of Things and deep learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月24日

查看全文
  • Weixi Liu

摘要

The growing integration of renewable energy into modern grids demands accurate wind power forecasting to ensure stable operations and efficient energy management. Existing methods often overlook nonlinear patterns and fail to fully exploit the Internet of Things (IoT)-based data acquisition, along with advanced deep learning-driven optimization for grid-connected power prediction and optimization. This research proposes a novel framework for grid-connected wind power prediction and optimization based on the IoT and deep learning. IoT-enabled sensors collect real-time meteorological and turbine operational data, which undergo enhanced preprocessing using the min-max normalization. Key predictive features are extracted using the Kernel Principal Component Analysis (Kernel-PCA) to capture nonlinear feature relationships and reduce dimensionality. The proposed Fennec Fox optimizer-tuned Dynamic Bidirectional Long Short-Term Memory with Attention Mechanism (FF-DBLSTM-AM) is employed to perform accurate grid-connected wind power prediction and optimization. The Fennec Fox optimizer ensures effective hyperparameter tuning, while the DBLSTM-AM architecture captures bidirectional temporal dependencies and applies attention to critical features for enhanced forecasting performance. The outputs are integrated into a grid connected optimization layer employing dynamic load balancing and energy dispatch strategies, enabling improved scheduling and reduced grid imbalance. Implemented in Python, the FF-DBLSTM-AM approach outperforms existing methods such as CNN-BIGRU, achieving an R-squared value of 0.9465(94.65%), RMSE of 0.1016, MAE of 0.0401, and MSE of 0.0112, demonstrating higher predictive accuracy and improved grid stability. These findings provide a reliable and scalable solution for enhancing wind energy integration on smart grids and highlight the advantages of the proposed model over conventional deep learning architectures.