非线性PCA故障检测方法及应用

Nonlinear PCA fault detection method and the application to petrochemical plant

针对化工过程的强非线性问题,提出一种基于神经网络的非线性主元分析故障检测方法,结合主元曲线算法和2个径向基神经网络,实现非线性主元的识别,并采用统计方法进行故障检测.第一个网络建立输入数据到非线性主元的映射,第二个网络实现逆映射重构原数据.在某炼油厂常压蒸馏过程的常压炉装置中的应用结果表明,基于神经网络的非线性主元分析故障检测方法的效果大大优于线性主元分析(PCA)方法,可准确进行故障检测和分离,保证常压炉安全高效地运行.

Fault detection in chemical process concerns much about the nonlinear problem,a nonlinear principal component analysis(NPCA) fault detection method based on neural networks is proposed,which combines two radial basis function(RBF) networks with a principal curves algorithm.The first network achieves the transformation of input variables to nonlinear principal components,and the desired output of the first network is obtained by the principal curves algorithm.The second network tries to perform the inverse transformation by reproducing the original data.The application to a distillation process in a petrochemical plant show the superiority compared to traditional linear principal component analysis(PCA) method,and precise fault detection and isolation results are achieved.