不同特征向量下基于SVM的短期风速预测

SHORT-TERM WIND SPEED FORECASTING BASED ON SVM UNDER DIFFERENT FEATURE VECTORS

选取广东省某风电场的测风数据,运用支持向量机(SVM)的方法对其进行短期风速预测。为提高预测的精度,通过LIBSVM回归机的交叉验证函数确定最优参数,建立4种不同输入特征向量组合(风速序列、风速和风向、风速和气压、风速风向和气压)的模型,分别预测该风场的短期风速,并对4种模型的预测误差进行分析和比较。实验结果表明:气压不宜作为输入特征向量;选用风速和风向作为输入特征向量的模型,预测效果最理想,其平均绝对百分比误差为12.8%。

The problem of short-term wind speed prediction is addressed based on support vector machine （SVM） using the data from a wind farm in Guangdong Province. In order to improve forecast accuracy, the cross- validation method of the LIBSVM software was adopted to determine the optimal parameters, and 4 types of models with 4 different feature vectors （wind speed, wind speed ＋ wind direction, wind speed ＋ air pressure, wind speed ＋ wind direction ＋ air pressure） were created to predict short-term wind speed of the wind field respectively, and the prediction errors were analyzed and compared. The experimental results showed that the pressure is not appropriate to be selected as feature variable, and the model with feature vectors of wind speed and wind direction takes the most ideal effect with the average absolute percentage error 12. 8%.