基于混沌相空间重构与支持向量机的风速预测  

WIND SPEED PREDICTION BASED ON CHAOS PHASE SPACE RECONSTRUCTION AND SUPPORT VECTOR MACHINE

作　　者：王富强[1] 王东风[1] 韩璞[1] 

机构地区：[1]华北电力大学控制与计算机工程学院,保定071003

出　　处：《太阳能学报》2012年第8期1321-1326,共6页Acta Energiae Solaris Sinica

基　　金：中央高校基本科研业务费专项资金(11MG49)

摘　　要：讨论了基于混沌相空间重构和支持向量机的风速预测。对风速数据的混沌特性进行分析,在此基础上进行相空间重构,采用C-C算法求取嵌入维数m和延时时间τ,确定预测模型的输入维m与样本集;在样本集中,采用粗搜索和细搜索的方法选取预测点的参考点,在进行细搜索的过程中提出相关性分离速率的方法,提高了预测精度。利用支持向量机强大的泛化能力,构造出风速预测回归函数,避免了传统的人工神经网络所存在收敛速度慢、结构选择困难和局部极小点问题。最后采用新西兰某风电场采样周期为10min的风速测量数据进行风速预测,实验结果表明采用Chaos-SVR方法有效降低了风速预测误差,且此方法与神经网络法相比具有更好的泛化能力和更高的预测精度。The wind speed prediction was discussed based on the chaos phase space reconstruction and the support vector machine. The chaotic characteristic of the wind data was analyzed, then the phase space reconstruction was taken. The embedded space dimension m and the delay time τ were calculated by the C-C algorithm. The reference points were taken out by rough search and fine search. In the fine search, the method of correlation forward rate was put out, so the accuracy of the predictions was improved. Using the generalization ability of the support vector machine （SVM）, the prediction regression function was constructed. In this way, the traditional neural network problems of the slow convergence speed, the structure selection and local minimum points were avoided. Finally, the New Zealand wind velocity measurement data were used by sampling period of 10min. The experimental results show that the prediction is effective by the Chaos-SVR and its generalization ability is better than neural network with higher precision of prediction.

关 键 词：混沌 相空间 支持向量回归 预测 

分 类 号：TK89[动力工程及工程热物理—流体机械及工程]