基于样本熵和支持向量机的短期风速预测模型

Forecasting model of short-term wind speed based on sample entropy and support vector machine

提出一种经验模态分解、样本熵和支持向量机相结合的短期风速组合预测方法。首先利用经验模态分解将原始风速序列逐级分解成若干个规律性更强的子序列,以减小不同特征尺度序列间的相互影响,提高预测精度。接着计算各风速子序列的样本熵,将复杂度相近的序列归类形成一个新序列,以减少所需建立的预测模型的数量。然后对经EMD-SE处理后得到的新的风速子序列分别建立支持向量机预测模型,并采用遗传算法实现各模型参数的自动选取和寻优,最后将各序列的预测结果叠加得到风速预测结果。算例研究表明,该方法充分挖掘了风速序列的特性,能快速地对风速变化作出响应,预测的均方根误差和百分比误差分别比单纯采用支持向量机法降低了5.1%和5.4%,有效地提高了短期风速预测的准确度。

This paper put forward a short-term wind speed forecasting method combined with empirical mode decomposition （EMD）, sample entropy （SE） and support vector machine （SVM）. Firstly, EMD was used to change the original wind speed sequence into several more regular subsequences step by step to minimize the mutual influence between different sequences and improve the prediction precision. Then calculating the SE of each wind speed sequence, cluster sequences of similar complexity was formed a new sequence, which can reduce the number of forecast model required. SVM prediction models was set up respectively for the new wind speed sequences operated by EMD-SE, and then automatic selection and optimization of model parameters can be realized by using genetic algorithm（GA）. Case study results showed that the method fully exploited the characteristics of the wind speed sequence, and can respond to the change of wind speed quickly. The RMSE and MAPE of prediction were reduced by 5.1 percent and 5.4 percent compared with using SVM separately, and the precision of shortterm wind speed forecasting was improved.