Support Vector Machine's Application in Significant Error Detection of Nonlinear Systems

Abstract

Presented a kind of principle and method based on regression support vector machine dynamic data significant error detection. The method takes full advantage of the nonlinear approximation capability supporting vector machine. The establishment of nonlinear system dynamic process model convex to a quadratic twice optimization problem, which can be guaranteed the extremal solution is global optimal solution and has good generalization ability. In this paper looked glutamic acid fermentation process as the research object, and established the chemical and biological variables prediction model based on SVM regression. At the same time achieved process variables online predicted. Through the method of strike the deviation of predicted value and measured to determine the existence of a significant error, which provide a new method for the significant error detection, eliminate and revise of dynamic process.