Scholarlink

<p>This study introduces a novel complete integrated framework to enhance the efficacy of short-term wind power forecasting, particularly for 15-min ahead predictions. The proposed hybrid model incorporates self-attention mechanism and combines the long short-term memory (LSTM) architecture with the eXtreme Gradient Boosting algorithm. In the context of data preprocessing, a multivariate approach was employed, aligning numerical weather prediction (NWP) wind speeds with the wind turbine hub height and incorporating data from the SCADA system for wind power. Empirical mode decomposition and principal component analysis are applied as feature engineering techniques to generate new features using NWP data. Bayesian optimization is used for hyperparameter tuning of LSTM model. Extensive experiments are carried out on datasets from four wind sites in Tamil Nadu, India, to evaluate the performance of the LAXGB model. The results validate the model’s effectiveness, accuracy, and robustness across diverse datasets and deep learning and feature extraction techniques.</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

Hybrid attention-based deep learning model using feature engineering approaches for wind power forecasting

  • 期刊论文

  • 发表日期：

    2025年8月27日

查看全文
  • Poonam Dhaka,
  • Mini Sreejeth,
  • M. M. Tripathi

摘要

This study introduces a novel complete integrated framework to enhance the efficacy of short-term wind power forecasting, particularly for 15-min ahead predictions. The proposed hybrid model incorporates self-attention mechanism and combines the long short-term memory (LSTM) architecture with the eXtreme Gradient Boosting algorithm. In the context of data preprocessing, a multivariate approach was employed, aligning numerical weather prediction (NWP) wind speeds with the wind turbine hub height and incorporating data from the SCADA system for wind power. Empirical mode decomposition and principal component analysis are applied as feature engineering techniques to generate new features using NWP data. Bayesian optimization is used for hyperparameter tuning of LSTM model. Extensive experiments are carried out on datasets from four wind sites in Tamil Nadu, India, to evaluate the performance of the LAXGB model. The results validate the model’s effectiveness, accuracy, and robustness across diverse datasets and deep learning and feature extraction techniques.