基于核主元分析和多级神经网络集成的汽轮机故障诊断  

Fault Diagnosis of Steam Turbine Based on Kernel Principal Component Analysis and Multistage Neural Network Ensemble

作　　者：李岩[1] 王东风[1] 韩璞[1] 

机构地区：[1]华北电力大学控制科学与工程学院,河北保定071003

出　　处：《电力科学与工程》2009年第6期67-71,共5页Electric Power Science and Engineering

基　　金：华北电力大学重大预研基金资助(20041306);华北电力大学留学回国人员科研启动基金资助(200814002)

摘　　要：提出一种基于核主元分析(KPCA)和多级神经网络集成的汽轮机故障诊断方法。该方法首先采用KPCA对汽轮机故障样本数据进行特征提取;然后计算相互独立训练出的多个神经网络个体在验证样本集上的泛化误差,并选择其中精确度较高的子神经网络作为集成的个体;最后采用基于正交最小二乘算法的径向基函数神经网络来集成各个子网的输出并得到最终的诊断结果。在某汽轮发电机组故障诊断中的应用表明,该方法具有较高的精确度和稳定性。One new method for fault diagnosis of steam turbine based on kernel principal component analysis （KPCA） and multistage neural network ensemble was proposed. Firstly, the sample data was analyzed using KPCA to extract main features. Then, the generalization error of the independently trained individual neural network to the validating set was calculated, according to which the individual neural networks whose generalization errors are in a threshold will be selected. Lastly, a radial basis function （RBF） neural network based on orthogonal least- squares （OLS）algorithm was used to combine the output of the selected member networks. The practical applications in fault diagnosis of steam turbine showed that the proposed approach gives promising results even with smaller learning samples, and it has higher accuracy and stability.

关 键 词：汽轮机 故障诊断 核主元分析 RBF神经网络 多级神经网络集成 

分 类 号：TK268[动力工程及工程热物理—动力机械及工程]