基于核主元分析和邻近支持向量机的汽轮机凝汽器过程监控和故障诊断

Process Monitoring and Fault Diagnosis of Condenser Using KPCA and PSVM

提出了基于核主元分析(KPCA)和邻近支持向量机(PSVM)的汽轮机凝汽器过程监控和故障诊断新方法,将数据先用核主元法进行分析和处理,即通过非线性变换将样本数据从输入空间映射到高维特征空间,然后在高维特征空间中进行特征提取,若数据的Hotelling’sT2和Q统计量超过控制限,说明有故障发生,则计算样本的非线性主元得分向量,并将其作为输入值送入已训练好的邻近支持向量机进行故障类型识别。该方法可以有效地捕捉变量间的非线性关系,过程监控和故障诊断效果明显好于PCA-PSVM法。汽轮机历史故障特征数据集仿真试验证明了该方法的有效性。

A new method for process monitoring and fault diagnosis of condenser based on kernel principle component analysis（KPCA） and proximal support vector machine （PSVM） was proposed. The data was first analyzed using KPCA, i.e., through a nonlinear mapping function, the data was projected from the input space to feature space and calculated Hotelling＇s T2 and Q statistics of the samples. If statistics value exceeded confidence limits of T2 and Q plots, it calculated nonlinear PC scores of fault data and then the nonlinear PC scores were fed into the well training multiple PSVMS to diagnose fault category. The proposed process monitoring and fault detection method could effectively capture the nonlinear relationship in process variables. Its performance significantly outperforms method based on PCA-PSVM. Simulating of the turbo generator fault data set proves that the method is effective.