【摘要】 为提高风电单机短期功率预测精度，提出一种以变分模态分解（VMD）为基础，分别构建长时间尺度趋势预测模块（LTTFM）和短时间尺度周期预测模块（STPFM）进行分类预测的风电单机短期功率预测模型。首先，针对风电功率序列本身的不平稳性，采用变分模态分解方法对原始功率序列进行分解，依据相关性将分解后各模态分为低频分量、高频分量以及残差。然后，根据各分量特性，设计不同的预测方法。基于门控循环单元（GRU）构建长时间尺度趋势预测模块对分解后的低频分量进行预测；基于维度变换与二维卷积神经网络（2D-CNN）构建短时间尺度周期预测模块对分解后的高频分量以及残差进行预测；最后，综合各分量预测结果获得最终的预测功率。采用风电场实际运行数据和公开数据集对提出的模型进行验证，结果表明所提模型预测误差（MAE）较常用模型下降34.9%～44.1%，具有较高的预测精度。更多还原

【Abstract】 To enhance the accuracy of short-term wind turbine power forecasting, a forecasting model based on variational mode decomposition(VMD) is proposed. The model constructs a long-term trend forecasting module(LTTFM) and a short-term periodic forecasting module(STPFM) for multi-time scale classification forecasting. Firstly, considering the non-stationarity of the wind power sequence, the original power sequence is decomposed by VMD, and the intrinsic mode functions are categorized into low-frequency components, high-frequency components, and residuals based on their correlation. Then, according to the characteristics of each component, different forecasting modules are designed. Long-term trend forecasting module based on GRU is constructed for the forecasting of low-frequency components. Short-term periodic forecasting module containing dimensional transformation and 2D-CNN is constructed for the forecasting of high-frequency components and residuals. Finally, the final power is obtained by synthesizing the forecasting results of each component. The proposed model is verified using actual operation data from a wind turbine, and the results show that the mean absolute error(MAE) of the proposed model is reduced by 34.9%-44.1% compared to commonly used models, indicating that the proposed model has a higher forecasting accuracy.更多还原