Scholarlink

<p>Accurate wind power prediction is crucial for enhancing the operational efficiency of power systems. To further improve prediction accuracy, a novel hybrid prediction model named TS-PatchTST is proposed in this paper. Initially, a wavelet transform is applied to decompose raw wind power sequences into multiple sub-components, extracting multi-scale data characteristics. Through a comparative analysis of five models, with and without a wavelet transform, confirms its effectiveness in boosting prediction accuracy. Furthermore, the Channel Independence Time Series Mixer (CI-TSMixer) backbone is integrated with the Patch Time Series Transformer (PatchTST) model, utilizing shared MLP to determine the correlations between different patches, effectively integrating information. Results demonstrate that the proposed model achieves superior accuracy, with MSE reduced to 0.08, MAE to 0.21, and MAPE to 0.17. Thus, the TS-PatchTST model, enhanced by a wavelet transform, offers significant advantages in wind power prediction.</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

Novel short-term wind power prediction hybrid model based on wavelet transform and TS-PatchTST

  • 期刊论文

  • 发表日期：

    2025年5月26日

查看全文
  • Mingju Gong,
  • Jiabin Huang,
  • Yining Wang,
  • Hanwen Cui,
  • Peng Yang,
  • Shaomin Jing,
  • Chen Dong

摘要

Accurate wind power prediction is crucial for enhancing the operational efficiency of power systems. To further improve prediction accuracy, a novel hybrid prediction model named TS-PatchTST is proposed in this paper. Initially, a wavelet transform is applied to decompose raw wind power sequences into multiple sub-components, extracting multi-scale data characteristics. Through a comparative analysis of five models, with and without a wavelet transform, confirms its effectiveness in boosting prediction accuracy. Furthermore, the Channel Independence Time Series Mixer (CI-TSMixer) backbone is integrated with the Patch Time Series Transformer (PatchTST) model, utilizing shared MLP to determine the correlations between different patches, effectively integrating information. Results demonstrate that the proposed model achieves superior accuracy, with MSE reduced to 0.08, MAE to 0.21, and MAPE to 0.17. Thus, the TS-PatchTST model, enhanced by a wavelet transform, offers significant advantages in wind power prediction.