基于多特征信息融合的风电机组整机性能评估

Performance assessment of wind turbine based on multi-feature information fusion

为解决风电机组单一健康状态模型可能发生误报的问题,提出一种基于信息融合的风电机组整机性能评估方法。首先改进了局部离群因子算法(LOF),用于筛选正常运行数据,使用Kendall相关系数进行参数选择,并基于深度信念网络(DBN)建立多个健康状态模型,提取实际运行数据与模型预测值的残差作为性能特征。再使用自组织映射神经网络(SOM)将残差空间映射到风电机组运行状态空间以实现信息融合,通过计算状态劣化指数来构建性能指标的方法,对风电机组进行性能评估。最后,通过实际的风电机组运行数据验证了所提方法的有效性。

To solve the problem that single-index normal behavior models of wind turbines might give false alarms,a new performance evaluation method based on information fusion for wind turbines was proposed.The Local Outlier Factor(LOF)algorithm was improved for screening normal operating data.Afterwards Kendall correlation coefficient was applied to parameter selection and several normal behavior models were constructed with Deep Belief Network(DBN),where the residuals between actual operating data and model prediction value represented performance features.Information fusion was then realized by mapping the residual space to operating behavior space using Self-organizing Mapping(SOM)neural network.On this basis,a performance index based on the deterioration degree of operating states was calculated for evaluating the performance of wind turbine.The validity and superiority of the proposed method was illustrated using real-world wind turbine operating data.