Font Size: a A A

The Radial Basis Function Neural Network Theory And Its Application To Desulfuration Intelligent Control System

Posted on:2003-01-02Degree:MasterType:Thesis
Country:ChinaCandidate:X L ZhangFull Text:PDF
GTID:2168360092965790Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
This paper discusses the theory about Neural Network,especially RBFNN, and the process of building Pangang hot metal desulfuration intelligent control model. A mature modeling technique in allusion to Pangang steel-making desulfuration pretreatment is still not in existence. After referring to domestic and overseas status quo, and studying and analyzing the desulfuration pretreatment technics and variety of factors, a modeling method of neural network based on radial basis function is elaborated in the paper,which can control the process of desulfuration pretreatment. According to the data available,the method can predict the operation parameter, and direct the production effectively, and improve the desulphuriaztion efficiency , yield and quality of finished product. This paper bring forwards a method that train RBFNN with the Recursive Orthogonal Least Squares (ROLS) algorithm after lucubrating the theory of neural network and the traditional algrithm of Orthogonal Least Squares,and select the ceters of network sequently using the information available in an ROLS algorithm after network thaining, that's backward selection algorithm.Theory and experiment prove that it require less computational space and faster speed than batch process using ROLS algorithm. It is shown that using backward selection algorithm select centers can minimize the net work output error in great extent , make network structure more simple and more generalization and approximation.Furthermoure,the offline control to Pangang hot metal desulfuration process is realized using RBFNN model. Campared with statistical analyze,it is shown that,the network structure and network output after trained RBFNN using improved ROLS is more reasonable than k-mean algrithm,and the control model has the property of self_learning,self_ organization and self_adaptive,and the control precision can be more than 90%.On the other hand,this paper also shows that,RBFNN model can control the desulfuration process on the whole in time,and the prediction result using RBFNN model is better than statistical analyze method.
Keywords/Search Tags:Recursive Orthogonal Least Squares (ROLS) algorithm, Radial Basis Function Neural Network (RBFNN), backward selection, generalization, approximation
PDF Full Text Request
Related items