Scholarlink

<p>With continuous advancements in wind power forecasting technology, prediction accuracy has progressively improved. However, the increasing scale and complexity of models present challenges for deployment on resource-limited devices. To address this issue, we propose a dual-layer convolutional DSC-GPC model designed to achieve high forecasting accuracy with a reduced parameter count in wind power forecasting. First, a missing value compensation method is used to impute missing data, while Multi-Variate Variational Mode Decomposition (MVMD) is applied to classify data into high and low modal energy categories. The data is then input into a dual-layer convolutional model: the first layer employs Depthwise Separable Convolution (DSConv) to independently capture each univariate dependency, while the second layer applies Grouped Pointwise Convolution (GPConv) to capture cross-variable dependencies among features. Finally, the overall prediction is obtained by combining the forecasting results of high and low modal energy data. Results indicate that the proposed model outperforms comparative models across multiple error metrics, with a parameter reduction of 45.3% compared to the GRU model, achieving the goal of a compact and efficient model with superior predictive accuracy.</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 lightweight dual-layer convolutional model for wind power forecasting

  • 期刊论文

  • 发表日期：

    2025年2月12日

查看全文
  • Yongsheng Ye,
  • Liu Liu,
  • Yang Li,
  • Xun Liu,
  • Yanlong Xu,
  • Yuchen Liu,
  • Lili Li

摘要

With continuous advancements in wind power forecasting technology, prediction accuracy has progressively improved. However, the increasing scale and complexity of models present challenges for deployment on resource-limited devices. To address this issue, we propose a dual-layer convolutional DSC-GPC model designed to achieve high forecasting accuracy with a reduced parameter count in wind power forecasting. First, a missing value compensation method is used to impute missing data, while Multi-Variate Variational Mode Decomposition (MVMD) is applied to classify data into high and low modal energy categories. The data is then input into a dual-layer convolutional model: the first layer employs Depthwise Separable Convolution (DSConv) to independently capture each univariate dependency, while the second layer applies Grouped Pointwise Convolution (GPConv) to capture cross-variable dependencies among features. Finally, the overall prediction is obtained by combining the forecasting results of high and low modal energy data. Results indicate that the proposed model outperforms comparative models across multiple error metrics, with a parameter reduction of 45.3% compared to the GRU model, achieving the goal of a compact and efficient model with superior predictive accuracy.