基于CEEMDAN和CS算法优化SVM的混合风速预测

Hybrid SVM with CEEMDAN and CS algorithm for wind speed prediction

基于自适应噪声完备集合经验模态分解(CEEMDAN)、布谷鸟算法(CS)和支持向量机(SVM)构建了CEEMDAN-CS-SVM混合风速预测模型,实现了黄土高原陇东区风电场月平均风速的准确预测.首先,采用CEEMDAN算法对收集到的风速时间序列进行去噪,以避免直接采用收集到的风速数据进行预测将导致较大误差的缺陷;其次,采用布谷鸟算法对SVM的惩罚系数和核函数半径进行优化,以克服SVM参数选择敏感的缺陷;最后,用构建的CEEMDAN-CS-SVM混合风速预测模型实现了黄土高原陇东区风电场月平均风速的预测.数值结果表明混合风速预测模型CEEMDAN-CS-SVM能够实现研究区域短期风速的准确预测,预测精度比混合模型DWT-SVM、EEMD-SVM、CEEMDAN-SVM、CS-SVM、DWT-CS-SVM、EEMD-CS-SVM及SVM的预测精度高.

A hybrid CEEMDAN-CS-SVM model was proposed based on the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),cuckoo search(CS)algorithm,and support vector machine(SVM)to predict the monthly average wind speed of wind farmer in Longdong area of the Loess Plateau,China.Firstly,CEEMDAN was employed to eliminate chaotic noise from the collected wind speed time series in order to counterbalance the weakness of forecasting wind speed using the collected wind speed time series directly may lead to large errors.Secondly,CS algorithm was adopted to optimize the penalty coefficient and kernel radius of SVM to overcome the defect of sensitive parameter selection.Finally,the proposed hybrid CEEMDAN-CS-SVM model was employed to predict the monthly average wind speed of wind farmer in Longdong area of the Loess Plateau,China.Numerical results show that CEEMDAN-CS-SVM model can accurately predict the monthly average wind speed in the study area,and the prediction accuracy is higher than that of hybrid DWT-SVM,EEMD-SVM,CEEMDAN-SVM,CS-SVM,DWT-CS-SVM,EEMD-CS-SVM and SVM.