基于高斯过程回归和双滑动窗口残差处理的风电机组主轴状态监测

Wind turbine spindle state monitoring based on Gaussian process regression and double moving window residual processing

针对风电机组运行数据强随机性和高噪声的特点,采用高斯过程回归方法建立主轴正常时段的温度模型,并应用自动关联分析方法选择主轴温度模型的建模变量。为降低建模复杂程度,采用模糊核聚类方法对风电机组原始运行数据进行筛选,消除冗余信息,构造紧凑有效的建模样本集。当主轴发生故障时,模型的输入观测向量发生异常变化,导致模型预测残差发生明显改变。为提高主轴异常预警的灵敏度和可靠性,采用基于莱依特准则的双滑动窗口对预测残差序列进行实时的统计分析,如果残差均值或标准差超出设定的故障报警阈值,则发出报警信息。某风电机组主轴的实际运行数据验证了所提方法的有效性。

Aiming at the strong randomness and high noise of the wind turbine operating data,the temperature model of the normal wind turbine spindle is established by the Gaussian process regression method,the input variables of which are selected by ARD（Automatic Relevance Determination） method.In order to reduce the complexity of modeling,the fuzzy kernel clustering method is used to filter the original operating data of wind turbine,and the redundant information is eliminated to construct a compact and effective modeling sample set.When the wind turbine spindle malfunctions,the input variables may abnormally change,which leads to the obvious variation of predicted residuals of the model.In order to improve the sensitivity and reliability of the early warning of spindle abnormity,the double moving window based on the Lewitt criteria is used to calculate the statistic properties of the residual sequence.If the mean value or standard deviation of residual exceeds the set fault alarm threshold,an alarm message will be issued.The actual operating data of wind turbine spindle verify the validity of the proposed method.