The Soft-Sensing Of Formaldehyde Concentration Based On SVM

In the real produce progress of industry, there are lots of progress parameters that can't be measure online. For example, in the produce progress of formaldehyde, as the lasted production, the concentration measurement of the formaldehyde, in the progress of biology ferment, the concentration measurement of 1, 3-propanediol and so on. In order to solve these kinds of problem, the soft-sensing technology is used widely. Soft-sensing usually gets the target parameter indirectly by detect, transform and calculation the fathomable parameter. The back promulgate network with at lest single hide layer can approach the nonlinear function at any precision, so there are many of scholars use Artificial Nerve Network to soft-sensing. But the ANN technology has some defections, first of all, the structure of the network is hard to fix; secondly, ANN easy to gets in local optimization point. Support Vector Machine is a learning technology based on structure risk minimization which promoted by Vapnik, its holds the strict theory and mathematics basement which doesn't exists local optimization, and doesn't rely on the quality and the number of training data excessively, so it fit to the complex industry progress of soft-sensing.Formaldehyde is widely used in vinylite, plastic, rubber, paper making, pharmacy, and antisepsis and so on. In the produce progress of formaldehyde, the concentration of formaldehyde is measured by titration way. So it is unavoidable that the persons touch the formaldehyde. But formaldehyde is virulent, that must do a great harm to the persons. In order to solve the important problem, this paper builds a soft-sensing model of formaldehyde concentration based on SVM. And in order to solve the key technology of the SVM in the real application, it promotes a parameter choose method based on quadratic programming, and validate its availability by simulation.First of all, this paper expounds the project background, soft-sensing technology and project implement of soft-sensing, expounds the modeling method and development state of soft-sensing. Secondly, to solve the problem of soft sensing of formaldehyde concentration, this paper builds a soft-sensing model of formaldehyde concentration using SVM, based on introduces the basic theory of Statistical Learning Theory and SVM. Thirdly, in order to identify the uncertain parameters, it presents a new strategy to select the parameters of the model by improving the cross validation method and grid-searching method using quadratic technology, there is a detailed arithmetic steps of the parameter choose method in the paper, and it easy to realize. At last, it presents the detail realization steps of the formaldehyde concentration soft-sensing, it validates the model based on SVM holds better extend ability than the model based on RBF, at the same time, it also validates the validity of the parameter choose method promoted by this paper is better than the method we usually used before.