基于时序注意力机制的超短期风电功率概率预测

Probability Prediction of Ultra-short-term Wind Power Based on Temporal Pattern Attention

提高预测精度是风电概率预测研究的关键问题,融合多源数值天气预报数据降低预测误差,采用时序注意力机制对输入信息进行自适应选择,采用时序卷积网络提取多时间尺度的概率特征,并使用混合Beta分布构建预测概率信息。算例结果表明通过时序注意力机制融合多源气象信息能有效提高模型训练的收敛性,其预测结果具有更高的精度。

Improving the prediction accuracy is a key problem for wind power probability prediction research.The multi-source numerical weather prediction were integrated to reduce the prediction error,the temporal pattern attention was used to select the input information adaptively,the temporal convolutional network was used to extract the multi-time scale probability features,and the mixed Beta distribution was used to construct the prediction probability information.The simulation results show that the convergence of model training can be improved effectively by integrating multi-source numerical weather prediction with temporal pattern attention,and the prediction results have higher accuracy.