基于SVR增量学习算法的变桨距风力机系统在线辨识

针对变桨距风力机模型非线性很强的特点,采用支持向量回归(SVR)算法进行辨识,数据由BLADED仿真软件提供,经训练检测的结果表明,SVR算法在变桨距风力机非线性模型辨识上具有很高的准确性。考虑到风力机现场工作过程中会出现模型变化,利用增量学习算法实现在线辨识。由于在线SVR辨识计算时间太长,通过改进的序列最小优化(SMO)算法代替原来的凸二次规划(QP)算法。同时提出满足度系数,排除系统无效的突变点,使在线辨识具有鲁棒性,并通过双支持向量机(SVM)算法实现在线辨识的记忆功能,最终辨识结果不仅有很强的精度,而且大大减小了计算时间。

INCREMENTAL LEARNING WITH SUPPORT VECTOR REGRESSION FOR PITCH-CONTROLLED WIND TURBINE ONLINE IDENTIFICATION

Support vector regression(SVR) was used for pitch-controlled wind turbine system identification,which is greatly nonlinear.The data was afforded by BLADED simulated software.After training,the testing result showed high veracity.But wind turbine's model may be changed under fieldwork.Therefore incremental learning algorithm is adopted for SVR online identification. However its calculation time is too long,so the improved sequential minimal optimization(SMO) algorithm was used to substitute the original quadratic programming(QP) algorithm.The satisfying coefficient was presented to eliminate the invalid break points,which made online identification robust.Moreover,the two support vector machine(SVM) algorithm made online identification having memory function.The final result indicated that the identification algorithm has high precision with short calculation time.