Scholarlink

<p>Reliable ultra-short-term wind speed forecasts support the operational performance, stability, and sustainable integration of wind energy systems. This study proposes a hybrid strategy combining variational mode decomposition, Bayesian optimisation, and a gated recurrent unit neural network. Variational mode decomposition separates wind speed into smoother frequency-dependent components, reducing non-stationarity and revealing dominant temporal patterns. These components are combined with multivariate supervisory control and data acquisition inputs and processed by the recurrent model, while Bayesian optimisation tunes its hyperparameters. The framework is evaluated for 10-, 20-, and 30-minute horizons and compared with Transformer, temporal convolutional network, long short-term memory, gated recurrent unit, and decomposition-based hybrid models. It achieves the lowest root mean square error and mean absolute error and the highest coefficient of determination. Wilcoxon tests with Holm correction confirm statistically significant improvements over the strongest competing model, while maintaining computational feasibility within the considered dataset and experimental protocol.</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 Ultra-Short-Term Wind Speed Forecasting using Variational Mode Decomposition and a Bayesian-Optimized GRU Model

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Sanae Arkhouch,
  • Ikrame Arrassi,
  • Ilyass El Myasse,
  • Nabil Benaya,
  • Faiza Dib,
  • Khaddouj Ben Meziane

摘要

Reliable ultra-short-term wind speed forecasts support the operational performance, stability, and sustainable integration of wind energy systems. This study proposes a hybrid strategy combining variational mode decomposition, Bayesian optimisation, and a gated recurrent unit neural network. Variational mode decomposition separates wind speed into smoother frequency-dependent components, reducing non-stationarity and revealing dominant temporal patterns. These components are combined with multivariate supervisory control and data acquisition inputs and processed by the recurrent model, while Bayesian optimisation tunes its hyperparameters. The framework is evaluated for 10-, 20-, and 30-minute horizons and compared with Transformer, temporal convolutional network, long short-term memory, gated recurrent unit, and decomposition-based hybrid models. It achieves the lowest root mean square error and mean absolute error and the highest coefficient of determination. Wilcoxon tests with Holm correction confirm statistically significant improvements over the strongest competing model, while maintaining computational feasibility within the considered dataset and experimental protocol.