基于核函数与概率神经网络的TE过程监控研究.pdfVIP

Proceedings of the 26th Chinese Control Conference July 26-31, 2007, Zhangji aj ie, Hunan, China 于核函数和概率神经网络的TE 过程监控研究* 薄翠梅，李 俊，陆爱晶，张广明 京工业大学自动化学院, 南京 210009 E-mail: lj_bcm@163.com 要：核主元分析方法是利用核函数计算高维特征空间的主元成分，从而把输入空间中的非线性问题转化为特征空 间中的线性问题。虽然核主元分析方法在故障检测方面明显优越于线性PCA，但故障辨识问题仍一个难题。本文提出 了一种基于核函数梯度算法和概率神经网络的新型非线性故障诊断方法。首先利用核函数梯度算法进行故障特征提 取，然后采用概率神经网络进行故障模式识 ，完成故障辨识过程。最后以Tennessee Eastman (TE) 化工过程的 15 种故障模式为例, 通过仿真实验证明KPCA-PNN 方法可保持较高的故障诊断率，体现了该网络具有令人满意的故障 诊断能力。 关键词：过程监控，核函数梯度算法，概率神经网络，Tennessee Eastman 过程 Nonlinear Process Monitors Method Based on Kernel Function and PNN Bo Cuimei, Li Jun, Lu Aij ing, Zhuang guangming Department of Automation Nanji ng University of Technology, Nanji ng, 2 10009, P. R. China E-mail: lj_bcm@163.com Abstract: Kernel PCA can efficiently compute principal components in high-dimensional feature spaces by means of nonlin- ear kernel functions. Therefore, the nonlinear problems are translated into the linear ones in the space high-dimension feature space. Although it has been proved that KPCA is superior to linear PCA for fault detection, the problem of fault identifica- tion theoretically has yet been a puzzle. A new fault detection and identification method based on the gradient arithmetic of kernel function and probabilistic neural network (PNN) for nonlinear system is developed. The gradient arithmetic of kernel function is used to extract the main features of faults firstly. Then, probabilistic neural network is used to identify the fault variables. To demonstrate the performance, the proposed method is applied to Tennessee Eastman processes. The simulation results