Scholarlink

<p>Wind power forecasting is essential for maintaining the stability and efficiency of electrical grids in the face of wind variability. This systematic review analyzes more than 100 recent publications and classifies artificial-intelligence (AI)-based wind power forecasting methods into five paradigms: classical, statistical, machine learning, deep learning, and hybrid. We organize and compare these approaches across four operational forecasting horizons—ultra-short-term (less than 1 hour), short-term (from 1 to 24 hours), medium-term (from 24 hours to 7 days), and long–term (more than 7 days)—using consistent metrics (root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and the coefficient of determination (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\varvec{R}^{\varvec{2}}\)</EquationSource> </InlineEquation>)). By combining a model-family taxonomy with a horizon–driven analysis, the review enables cross-paradigm insights into performance trends, data requirements, and practical constraints. We also discuss key challenges, notably the growing complexity of hybrid architectures and their associated computational costs. Finally, we offer recommendations to standardize evaluation protocols and guide future research toward lighter, more robust solutions.</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

A Comprehensive Review of AI-based Wind Power Forecasting Over Multiple Time Horizons

  • 期刊论文

  • 发表日期：

    2026年3月14日

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

摘要

Wind power forecasting is essential for maintaining the stability and efficiency of electrical grids in the face of wind variability. This systematic review analyzes more than 100 recent publications and classifies artificial-intelligence (AI)-based wind power forecasting methods into five paradigms: classical, statistical, machine learning, deep learning, and hybrid. We organize and compare these approaches across four operational forecasting horizons—ultra-short-term (less than 1 hour), short-term (from 1 to 24 hours), medium-term (from 24 hours to 7 days), and long–term (more than 7 days)—using consistent metrics (root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and the coefficient of determination ( \(\varvec{R}^{\varvec{2}}\) )). By combining a model-family taxonomy with a horizon–driven analysis, the review enables cross-paradigm insights into performance trends, data requirements, and practical constraints. We also discuss key challenges, notably the growing complexity of hybrid architectures and their associated computational costs. Finally, we offer recommendations to standardize evaluation protocols and guide future research toward lighter, more robust solutions.