Scholarlink

<p>Accurate short-term wind power forecasting (STWPF) is essential for grid operators and energy traders, particularly for managing grid stability and supporting utility energy dispatch in the short term. This research introduces a hybrid approaches that integrates multi-feature wavelet decomposition (MFWD) technique with convolution-gated recurrent unit (CGRU) to improve forecasting accuracy. The model is validated using historical data from the supervisory control and data acquisition (SCADA) system of the Kelmarsh wind farm (KWF). The MFWD pre-processing technique is employed to denoise the input features by decomposing each feature into detailed (high-frequency) and approximate (low-frequency) coefficients. This decomposition improves feature quality and enhances their Pearson's correlation coefficient (PCC), ensuring more relevant and robust inputs. The pre-processed dataset is then fed into the CGRU model, where convolutional layers extract deep spatial features, and GRU layers effectively capture temporal dependencies. The results demonstrate that the proposed hybrid MFWD–CGRU approach significantly outperforms the conventional CGRU model by 3.5% to 6.1% in multi-step forecasts horizons ranging from 1 to 24&#xa0;h ahead. The proposed model effectively reduces noise, improves feature relevance, and enhances overall forecasting accuracy. These improvements support better decision-making, strengthen grid reliability, and boost operational effectiveness.</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

Multi-step ahead wind power forecasting based on multi-feature wavelet decomposition and convolution-gated recurrent unit model

  • 期刊论文

  • 发表日期：

    2025年3月12日

查看全文
  • Shubham Shringi,
  • Lalit Mohan Saini,
  • Sanjeev Kumar Aggarwal

摘要

Accurate short-term wind power forecasting (STWPF) is essential for grid operators and energy traders, particularly for managing grid stability and supporting utility energy dispatch in the short term. This research introduces a hybrid approaches that integrates multi-feature wavelet decomposition (MFWD) technique with convolution-gated recurrent unit (CGRU) to improve forecasting accuracy. The model is validated using historical data from the supervisory control and data acquisition (SCADA) system of the Kelmarsh wind farm (KWF). The MFWD pre-processing technique is employed to denoise the input features by decomposing each feature into detailed (high-frequency) and approximate (low-frequency) coefficients. This decomposition improves feature quality and enhances their Pearson's correlation coefficient (PCC), ensuring more relevant and robust inputs. The pre-processed dataset is then fed into the CGRU model, where convolutional layers extract deep spatial features, and GRU layers effectively capture temporal dependencies. The results demonstrate that the proposed hybrid MFWD–CGRU approach significantly outperforms the conventional CGRU model by 3.5% to 6.1% in multi-step forecasts horizons ranging from 1 to 24 h ahead. The proposed model effectively reduces noise, improves feature relevance, and enhances overall forecasting accuracy. These improvements support better decision-making, strengthen grid reliability, and boost operational effectiveness.