基于属性约简重构的自校正卷积记忆风速预测

Wind Speed Prediction With Self-tuning Convolutional Memory Based on Attribute Reduction Reconstruction

考虑风速属性及时空相关性的预测建模是规模化风电并网的研究热点,该文基于属性约简重构提出一种自校正卷积记忆超短期预测模型。利用快速相关性滤波对风速序列关联属性进行排序筛选,据此改进K-mediods方法对风电场机群聚类,基于改进灰色关联度分析簇内风机的风速时空相关性,划分典型风机多阶邻域,并重构风速信息矩阵。然后,将重构的时空多维信息输入卷积双层记忆网络,通过卷积神经网络进行风速信息降维与空间特征提取,再由双层记忆神经网络进行多位置多步超短期预测,同时基于反向误差传播原理在记忆网络中引入自校正误差修正单元。最后对实际风电场的风速进行预测,验证所提方法的有效性。

Predictive modeling that considers wind speed attributes and temporal-spatial correlation is a research hotspot in large-scale wind power grid integration.This paper proposes a self-tuning convolutional memory ultra-short-term prediction model based on attribute reduction reconstruction.Use fast correlation filtering to sort and filter the correlation attributes of wind speed series,and then improve the K-mediods method to cluster wind farm clusters,analyze the temporal and spatial correlation of wind speeds in clusters based on improved gray correlation,and divide typical wind turbines into multi-level neighborhoods,And reconstruct the wind speed information matrix.Then,the reconstructed space-time multi-dimensional information is input into the convolutional double-layer memory network,wind speed information is reduced by the convolutional neural network,and spatial feature extraction,and then the double-layer memory neural network is used for multi-position and multi-step ultra-short-term prediction.The principle of reverse error propagation introduces a self-correcting error correction unit in the memory network.Finally,the wind speed of the actual wind farm is predicted to verify the effectiveness of the method in this paper.