Scholarlink

To address the limitation that conventional multi-dimensional wind power prediction datasets incorporating meteorological information often fail to achieve satisfactory forecasting accuracy, this study proposes an ultra-short-term wind power forecasting framework integrating Kernel Principal Component Analysis (KPCA) with an Attention-based Long Short-Term Memory (LSTM) network. The proposed methodology employs KPCA for nonlinear dimensionality reduction of all features in the wind farm dataset, effectively extracting the most salient characteristics while eliminating redundant information. Furthermore, an attention mechanism is incorporated to enable the LSTM network to dynamically allocate higher weights to the feature elements that exert greater influence on the target prediction moment. Experimental validation using actual historical operational data from wind farms demonstrates that the proposed model significantly enhances ultra-short-term wind power forecasting performance, outperforming both traditional Back Propagation Neural Networks (BPNN) and conventional LSTM architectures in terms of prediction accuracy.

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

Attention-LSTM Wind Power Ultra-Short-Term Prediction Based on Kernel Principal Component Analysis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lihao Zheng,
  • Jiandong Duan,
  • Pengfei Zhang,
  • Tiancheng Mu,
  • Yaoyun Zhang

摘要

To address the limitation that conventional multi-dimensional wind power prediction datasets incorporating meteorological information often fail to achieve satisfactory forecasting accuracy, this study proposes an ultra-short-term wind power forecasting framework integrating Kernel Principal Component Analysis (KPCA) with an Attention-based Long Short-Term Memory (LSTM) network. The proposed methodology employs KPCA for nonlinear dimensionality reduction of all features in the wind farm dataset, effectively extracting the most salient characteristics while eliminating redundant information. Furthermore, an attention mechanism is incorporated to enable the LSTM network to dynamically allocate higher weights to the feature elements that exert greater influence on the target prediction moment. Experimental validation using actual historical operational data from wind farms demonstrates that the proposed model significantly enhances ultra-short-term wind power forecasting performance, outperforming both traditional Back Propagation Neural Networks (BPNN) and conventional LSTM architectures in terms of prediction accuracy.