基于CEEMDAN-FE-TCN-GRU-AM的风电功率多步预测模型  

Wind Power Multi-Step Prediction Model Based on CEEMDAN-FE-TCN-GRU-AM

作　　者：刘世卓 张一梅 Shizhuo Liu;Yimei Zhang(College of Environmental Science and Engineering,North China Electric Power University,Beijing)

机构地区：[1]华北电力大学环境科学与工程学院,北京

出　　处：《建模与仿真》2025年第2期717-729,共13页Modeling and Simulation

摘　　要：超短期风电功率预测对电力系统的运行和并网具有重大意义,本文为提高风电功率预测精度,提出一种基于CEEMDAN-FE-TCN-GRU-AM的混合深度学习预测模型。首先利用四分位法对数据进行异常值处理,再利用皮尔逊系数(PCC),最大信息系数法(MIC)方法对特征进行筛选,减少冗余的特征维度,增强模型对特征的理解,然后采用CEEMDAN分解方法将非平稳的原始功率分解为多个子序列和残差,再计算每个子序列的模糊熵(FE)后利用K-means算法对原始序列进行重组得到高中低三个频率特征,最后将重组的频率特征与原有特征结合输入到TCN-GRU-AM模型中进行超短期多步预测。结果说明了该模型相较于基准模型,具有更高的预测精度。Ultra-short-term wind power prediction is of great significance to the operation and grid connection of power systems.This paper proposes a hybrid deep learning prediction model based on CEEMDAN-FE-TCN-GRU-AM.First,the quartile method is used to process the outliers of the data,and then the Pearson coefficient(PCC)and the maximum information coefficient method(MIC)are used to screen the features to reduce the redundant feature dimensions and enhance the model’s understanding of the features.Then,the CEEMDAN decomposition method is used to decompose the nonstationary original power into multiple subsequences and residuals.After calculating the fuzzy entropy of each subsequence,the K-means algorithm is used to reorganize the original sequence to obtain three frequency features of high,medium and low.Finally,the reorganized frequency features are combined with the original features and input into the TCN-GRU-AM model for ultra-shortterm multi-step prediction.The results show that the model has higher prediction accuracy than the benchmark model.

关 键 词：风电功率预测 CEEMDAN分解 模糊熵 GRU 注意力机制 

分 类 号：TM614[电气工程—电力系统及自动化]