Abstract: Support vector machines generalized observer (SGO) was proposed and its design was introduced. The observer employs support vector machines (SVM) regression algorithm for fitting the nonlinearity among process variables. It was derived by process inputs and outputs except for the output to be monitored and can be used for process fault detection and fault tolerance control. multi-variables chemical process simulation results show that SVM overcomes some disadvantages of artificial neural network (ANN), such as over fitting, local minimization and difficulties in structure selection. SGO has a quite precise output in application.

Key words: Support vector machines , Regression algorithm , Generalized observer , Artificial neural network 

摘要：

提出了广义支持向量机观测器的概念并介绍了其设计方法。该观测器采用支持向量机回归算法拟
合过程变量之间的非线性关系, 由过程输入和除被观测输出之外的其它过程输出进行驱动, 可用于实现过程传感器
的故障检测和容错控制。多变量化工过程仿真实验表明, 广义支持向量机观测器克服了神经网络类方法在应用时
所存在的过学习、易陷入局部极小和结构选择困难等缺陷, 并且达到了很高的拟合精度。

关键词: 支持向量机 , 回归算法 , 广义观测器 , 神经网络