综合智慧能源

• •    

融合CEEMDAN-CNN-LSTM的风电机组多气象场景功率回归预测

皇甫陈萌, 阮贺彬, 徐俊俊   

  1. 南京信息工程大学自动化学院, 江苏 210044 中国
    南京邮电大学自动化学院、人工智能学院, 江苏 210023 中国
  • 收稿日期:2025-06-09 修回日期:2025-07-16
  • 基金资助:
    国家自然科学基金项目(52107101)

CEEMDAN-CNN-LSTM-Based Power Regression Prediction for Wind Turbines in Diverse Meteorological Scenarios

  1. , 210044, China
    , 210023, China
  • Received:2025-06-09 Revised:2025-07-16
  • Supported by:
    National Natural Science Foundation of China(52107101)

摘要: 为提高不同气象场景下风电机组输出功率预测的准确性,提出了一种基于自适应噪声完备集合经验模态分解-卷积神经网络-长短期记忆网络模型的风电机组功率回归预测方法。首先,通过自适应噪声完备集合经验模态分解(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN)算法对原始风电功率数据进行分解,利用本征模态函数(Intrinsic Mode Functions,IMFs)和残差项(Residual,RES),并考虑风速等五种气象因素,结合卷积神经网络(Convolutional Neural Network,CNN)进行特征提取,采用长短期记忆网络(Long Short-Term Memory,LSTM)对每个子序列进行回归预测,并将预测结果进行叠加重构,得到最终预测值,使用平均绝对误差和均方根误差量化预测值与实际值之间的偏差。通过某地区实测数据对比分析,结果表明,该方法通过引入CEEMDAN和CNN对原始气象数据进行特征提取,相比现有预测方法,有效提高了特征捕捉能力,同时结合常规气象条件下的模型训练与迁移,进一步增强了模型对多气象场景及极端天气小样本数据的拟合和泛化能力。

关键词: CEEMDAN-CNN-LSTM模型, 风电机组, 功率预测, 气象因素, 极端天气

Abstract: To improve the accuracy of wind turbine power prediction under different meteorological scenarios, a method based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise - Convolutional Neural Network - Long Short-Term Memory is proposed. The original wind power data is decomposed into Intrinsic Mode Functions (IMFs) and Residual (RES) using CEEMDAN, considering five meteorological factors including wind speed. These factors were combined with a CNN for feature extraction. Each subsequence of wind power is predicted by LSTM, and the results are combined to obtain the final prediction. The prediction accuracy is evaluated using Mean Absolute Error and Root Mean Squared Error. Tests with real data from a region show that this method effectively improves feature extraction through CEEMDAN and CNN. Combined with model training and transfer under normal meteorological conditions, it enhances the model's fitting and generalization capabilities for diverse and extreme weather scenarios.

Key words: CEEMDAN-CNN-LSTM model, wind turbine, power prediction, meteorological factors, extreme weathers