EEMD和CNN-XGBoost在风电功率短期预测的应用研究

Application of EEMD and CNN-XGBoost in short-term wind power prediction

为解决风电功率序列随机性强、波动性大、预测误差高的问题,提出一种基于集合经验模态分解与卷积神经网络-极端梯度提升相结合的短期风电功率组合预测模型。该模型首先对原始的风电功率序列进行预处理,剔除缺失值和离群值;其次进行EEMD分解得到一系列子序列;再将每组子序列输入到CNN模型中提取特征信息;最后采用XGBoost回归模型对风电功率进行预测,并和XGBoost、CNN-XGBoost两种预测模型进行对比;经甘肃某风电场的实际风电运行数据验证,EEMD-CNN-XGBoost预测模型具有更好的预测效果以及更高的预测精度。

In order to solve the problems of strong randomness, large volatility and high prediction error of wind power series, a short-term wind power prediction model based on the combination of esemble empirical model decomposition(EEMD) and convolutional neural network-eXtreme gradient boosting(CNN-XGBoost) is proposed. Firstly, the wind power sequence is pretreated to eliminate missing values and outliers. Then perform esemble empirical model decomposition to get a series of subsequences. Then input each group of subsequences into the convolutional neural network to extract feature information. Finally, the XGBoost regression model is used to predict the wind power and compare with XGBoost and CNN-XGBoost models. The actual wind power operation data of a wind farm in Gansu province show that the EEMD-CNN-XGBoost prediction model has better prediction effect and higher prediction accuracy.