针对非平稳信号和高频噪声的自适应噪声完整集成经验模态分解-双向长短期记忆风功率预测模型  

An Improved CEEMDAN-BiLSTM Model for Wind Power Prediction Addressing Non-stationary Signals and High-frequency Noise

作　　者：万思洋 杨苹[1,2] 崔嘉雁 李丰能 隗知初 陈文皓 WAN Siyang;YANG Ping;CUI Jiayan;LI Fengneng;WEI Zhichu;CHEN Wenhao(School of Electric Power,South China University of Technology,Guangzhou 510640,Guangdong Province,China;Key Laboratory of Clean Energy Technology of Guangdong Province(South China University of Technology),Guangzhou 510640,Guangdong Province,China)

机构地区：[1]华南理工大学电力学院,广东省广州市510640 [2]广东省绿色能源技术重点实验室(华南理工大学),广东省广州市510640

出　　处：《电网技术》2025年第3期1176-1184,I0085,共10页Power System Technology

基　　金：广东省海洋经济发展(海洋六大产业)专项资金项目(GDNRC[2023]27)。

摘　　要：提出了一种基于改进的自适应噪声完整集成经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)和双向长短期记忆神经网络(bidirectional long short-term memory,BiLSTM)的组合预测模型,以提高风电功率预测的准确性和鲁棒性。当前风电功率预测面临非平稳信号和高频噪声的问题,影响了预测的准确性。针对这一问题,通过CEEMDAN分解,将复杂的非平稳信号分解为多个固有模态函数分量(intrinsic mode function,IMF),在此基础上创新性地通过平均波动幅度(average fluctuation range,AFR)计算IMF的平均波动幅度进行高低频划分,应用经验小波变换(empirical wavelet transform,EWT)对高频分量进行滤波,显著降低信号中的高频噪声,提高数据准确性。随后,分别对高频和低频分量建立Bi-LSTM模型,选取最优参数进行训练和预测,将各分量的预测结果叠加得到最终的风电功率预测值。模型经过不同季节和数据集的验证,展示了其在风电功率预测中的通用性和鲁棒性。研究证明,结合CEEMDAN分解、AFR划分和EWT滤波,通过有效的噪声抑制和数据分解,能够显著提升风电功率预测的准确性和稳定性,弥补了传统方法在处理非平稳信号和高频噪声方面的不足。This paper proposes a combined forecasting model based on improved CEEMDAN and BiLSTM to enhance wind power forecasting accuracy and robustness.The model uses CEEMDAN decomposition to break down complex signals into multiple IMFs to address challenges from non-stationary signals and high-frequency noise.Innovatively,AFR calculates the average fluctuation range of IMFs for frequency division,and EWT filters high-frequency components,significantly reducing noise.BiLSTM models are then trained on these components with optimal parameters.Validation across different seasons and datasets demonstrates the model's generality and robustness.This approach effectively improves prediction accuracy and stability,addressing traditional methods'limitations in handling non-stationary signals and noise.

关 键 词：风电功率预测 双向长短期记忆神经网络 完全集成经验模态分解 经验小波变换 深度学习 

分 类 号：TM614[电气工程—电力系统及自动化]