基于量子进化在线序贯极限学习机的变桨系统故障检测

针对复杂工况下风电机组变桨系统故障检测问题,采用在线序贯极限学习机建立变桨系统状态监测模型,利用ReliefF算法进行模型的特征选择,通过量子进化算法优化在线序贯极限学习机的超参数集,并引入马氏距离函数计算变桨系统状态监测模型的残差,判断风电机组变桨系统的异常。以辽宁某风电场1.5 MW双馈风电机组变桨系统为例,将所提出的模型分别与粒子群优化极限学习机、粒子群优化支持向量机、随机权神经网络、极限学习机和反向传播神经网络模型进行对比,结果表明所提出的模型精度优于其他模型,所提方法的故障检测正确率高于3σ阈值法和核主成分分析方法。

FAULT DETECTION BASED ON ONLINE SEQUENTIAL EXTREME LEARNING MACHINE USING QUANTUM EVOLUTIONARY ALGORITHM FOR PITCH SYSTEM

For the problem of pitch system fault detection of wind turbine generations under complex working conditions,an online sequential limit learning machine is employed to establish a condition monitoring model for the pitch system,the ReliefF algorithm is adopted to select the characteristic variables of the model,and the hyper parameter set of the online sequential extreme learning machine are optimized by quantum evolutionary algorithm.The Mahalanobis distance function is introduced to calculate the residual error of the condition monitoring model of the pitch system to detect the abnormality of the pitch system for wind turbine.Taking a pitch system of 1.5 MW doubly-fed wind turbine located in Liaoning Province as an example,the proposed model is compared with PSOELM,PSO-SVR,RVFLN,ELM,BPNN models.The results show that the proposed method has better accuracy than the others.And the fault detection accuracy is superior to the 3σthreshold and KPCA method.