基于信息调控和MATCN的超短期风电功率多步预测

Ultra⁃short⁃term wind power multi⁃step forecasting based on information regulation and MATCN

对波动的风电功率进行有效预测,是电网供需平衡、系统稳定运行的重要保障。为此,提出一种基于信息调控和MATCN的超短期风电功率多步预测方法。利用现有数据衍生出高阶项与交互项,提升特征序列数量与有效特征占比。针对复杂的风电数据结构,使用变分模态分解(VMD)将其拆分,根据子序列相关性和方差贡献率的计算结果保留重要序列分量,其余分量进行聚合,降低计算负担,缩短训练时间。随后,引入注意力机制构造多头注意力时间卷积网络(MATCN),通过注意力得分调整网络内部卷积单元之间的传递信息,实现模型对各序列分量的预测。最后,重构序列分量预测值,得到最终的输出结果。在实例数据上对所提模型进行对比验证,结果表明,该模型在不同步幅下均具有较好的预测效果。

The effective forecasting of fluctuating wind power is an important guarantee for the balance of power supply and demand and the stable operation of the system.Therefore,a method of ultra-short-term wind power multi-step forecasting based on information regulation and multi-head attention temporal convolution network(MATCN)is proposed.High-order items and interactive items are derived from the existing data to increase the proportion of the number of feature sequences and effective features.The variational mode decomposition(VMD)is used to split the complex wind power data structure,the important sequence components are retained according to the calculation results of sub-sequence correlation and variance contribution rate,and other components are aggregated to reduce the calculation burden,and shorter training time.attention mechanism is introduced to construct the MATCN,and the transmitted information between convolution units within the network is adjusted by the attention score,so as to realize the prediction of each sequence component of the model.The sequence component prediction values are reconstructed to obtain the final output result.The proposed model is compared and verified on example data,and the results show that the model has excellent prediction effects under different strides.