Scholarlink

<p>The increasing integration of renewable energy sources has heightened the need for accurate power forecasting. However, photovoltaic (PV) and wind power outputs remain highly volatile, and the complexity of environmental conditions poses significant challenges to hyperparameter tuning in predictive models. To address this, an improved sine algorithm and dung beetle optimization (MSADBO) is employed to automatically optimize model hyperparameters and enhance forecasting performance. A hybrid deep learning framework is proposed, which combines a bidirectional gated recurrent unit (BiGRU) for sequential modeling, a temporal convolutional network (TCN) for capturing long-range dependencies, and a self-attention mechanism to strengthen temporal feature extraction. These components are integrated into the proposed predictive model, named MSADBO-AT-BiGRU-TCN, with MSADBO used to optimize the architecture’s hyperparameters. Extensive experiments on real-world PV and wind datasets demonstrate that the proposed model consistently achieves lower prediction errors and greater robustness under abrupt power fluctuations. It outperforms both traditional deep learning baselines and recent optimization-based forecasting methods across multiple evaluation metrics. Furthermore, its modular and scalable design facilitates efficient deployment on high-performance computing platforms, enabling real-time, large-scale renewable energy forecasting.</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

Optimizing renewable energy forecasting: a hybrid approach integrating MSADBO, BiGRU, and TCN for PV/wind power generation prediction

  • 期刊论文

  • 发表日期：

    2025年8月23日

查看全文
  • Hang Wan,
  • Zhizhuo Qiu,
  • Jiasong Wang,
  • Rui Quan,
  • Yufang Chang,
  • William Derigent

摘要

The increasing integration of renewable energy sources has heightened the need for accurate power forecasting. However, photovoltaic (PV) and wind power outputs remain highly volatile, and the complexity of environmental conditions poses significant challenges to hyperparameter tuning in predictive models. To address this, an improved sine algorithm and dung beetle optimization (MSADBO) is employed to automatically optimize model hyperparameters and enhance forecasting performance. A hybrid deep learning framework is proposed, which combines a bidirectional gated recurrent unit (BiGRU) for sequential modeling, a temporal convolutional network (TCN) for capturing long-range dependencies, and a self-attention mechanism to strengthen temporal feature extraction. These components are integrated into the proposed predictive model, named MSADBO-AT-BiGRU-TCN, with MSADBO used to optimize the architecture’s hyperparameters. Extensive experiments on real-world PV and wind datasets demonstrate that the proposed model consistently achieves lower prediction errors and greater robustness under abrupt power fluctuations. It outperforms both traditional deep learning baselines and recent optimization-based forecasting methods across multiple evaluation metrics. Furthermore, its modular and scalable design facilitates efficient deployment on high-performance computing platforms, enabling real-time, large-scale renewable energy forecasting.