Scholarlink

<p>Accurate modeling of wind turbine power curves is crucial for evaluating turbine performance and improving wind power forecasting. To address the uncertainty in the wind speed–power relationship and the complexity of power curve distributions, this study proposes a multivariate modeling approach based on a Temporal Convolutional Network with Self-Attention (TCN-SA). The TCN utilizes causal and dilated convolutions to extract temporal features from wind speed and environmental variables, while a multi-head self-attention mechanism from the Transformer architecture integrates multi-factor information. This structure enhances the model’s ability to capture long-range dependencies and nonlinear relationships, improving robustness and fitting accuracy under complex conditions. Experiments using SCADA data from a wind farm in northwest China demonstrate that the proposed TCN-SA model significantly outperforms traditional methods. Across four turbines, the average MAE and RMSE are 29.9234 kW and 44.946 kW, respectively. Compared with cubic B-spline, 10th-order polynomial, and 1D CNN models, the proposed approach reduces MAE by 15.49–19.81% and RMSE by 12.52–16.80%. These results confirm the effectiveness of the proposed method for wind turbine power curve modeling and indirect 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

Wind Turbine Power Curve Modeling Method Based on TCN-SA Model

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Qiaoling Yang,
  • Binzhen Ma,
  • Kai Chen

摘要

Accurate modeling of wind turbine power curves is crucial for evaluating turbine performance and improving wind power forecasting. To address the uncertainty in the wind speed–power relationship and the complexity of power curve distributions, this study proposes a multivariate modeling approach based on a Temporal Convolutional Network with Self-Attention (TCN-SA). The TCN utilizes causal and dilated convolutions to extract temporal features from wind speed and environmental variables, while a multi-head self-attention mechanism from the Transformer architecture integrates multi-factor information. This structure enhances the model’s ability to capture long-range dependencies and nonlinear relationships, improving robustness and fitting accuracy under complex conditions. Experiments using SCADA data from a wind farm in northwest China demonstrate that the proposed TCN-SA model significantly outperforms traditional methods. Across four turbines, the average MAE and RMSE are 29.9234 kW and 44.946 kW, respectively. Compared with cubic B-spline, 10th-order polynomial, and 1D CNN models, the proposed approach reduces MAE by 15.49–19.81% and RMSE by 12.52–16.80%. These results confirm the effectiveness of the proposed method for wind turbine power curve modeling and indirect wind power prediction.