支持向量机向量维数对短期风电功率预测精度的影响

Influence of the vector-dimension of SVM on short-term wind power prediction accuracy

为了提高短期风电功率的预测精度,采用支持向量机回归方法,以10min采样间隔的实测风速和温度序列为输入数据,针对连续19天的历史数据按每隔5h进行一次预测,每次采用一天144组风速、温度数据进行训练。实验分别选取了10min、30min和1h的预测时间尺度,得到不同向量维数下预测结果的均方误差、均方百分比误差、平均绝对误差、平均绝对百分比误差以及相关系数。结果表明,对于特定的预测对象,支持向量机向量维数具有一个使预测精度最高、误差最小的最优值;对于不同的预测时间尺度,向量维数的最优值一般不同。

In order to improve prediction accuracy of short-term wind power forecasting, this paper presents a novel approach of Support Vector Regression （SVR） in short-term wind power forecasting by using measured temperatures and wind speed data of 10 minutes sampling interval. For historical data of 19 days, prediction is made once every five hours, and each time 144 groups of temperatures and wind speed data in a day are used for training. Experiments are carried out based on 10 min, 30 min and one hour time scales respectively. MSE, MSPE, MikE, MAPE and correlation coefficient are also calculated under varying vector-dimension. It can be seen that there is an optimal vector-dimension which has the highest accuracy and minimum error for specific prediction. Results show that for different time scales, the optimal value of the vector dimension is generally different.