Font Size: a A A

A Modified Radial Basis Function Neural Network Research And Application

Posted on:2004-12-23Degree:MasterType:Thesis
Country:ChinaCandidate:H Q WangFull Text:PDF
GTID:2168360095956796Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
This paper discusses the algorithm of RBFNN(Radial Basis Function Neural Network), especially modified RBFNN based on chaos genetic algorithm, and the modified algorithm application of building converter vanadium static model.Through referring to domestic and overseas research status, and studying and analyzing the several researches achievements of RBFNN and current development, to overcome the existent shortcoming and to resolve the practical engineering, a model method of modified RBFNN based on chaos genetic algorithm is put forward. It is used to distinguish reaction process of converter vanadium because presently, a mature and dependable mathematic model is still not in existence, which can control and guild converter vanadium process how to operate. According to test data from laboratory before distilling vanadium, the model in practice is used to predict how to control operation for ensuring effect of distilling vanadium and quality of middle-stage-steeling. After adopting the model, the goal of "distilling vanadium and simultaneously reserve carbon" is realized successfully.In accordance with requirement of building model, this paper researches and modifies building model and optimization method. At first, this paper in detailed analyzes the basic theory of entire searching optimized answer based genetic algorithm. Aim at gene evolvement limitation of traditional genetic algorithm, this paper presents a modified genetic algorithm which on the one hand adaptively rectifies crossover and mutation probability to improve gene evolvement effect, on the other hand applies chaos sequence to regularly disorder initialization of gene. After adopting these two methods, the occurrence of prematurity phenomenal is overcome. Simulation has proved the validity of chaos genetic algorithm.In succession, this paper deeply discusses neural network theory and traditional radial basis function algorithm. Based on this studying, for resolving RBFNN approximation and generalization, combination of RBFNN and chaos genetic algorithm is put forward, which make the most of RBFNN approximation and chaos genetic algorithm entire optimization. The joint of two algorithms is utilizing chaos genetic algorithm to resolve RBFNN connection node weight of hidden-output layer. Similarly simulation has testified the rationality of the modified RBFNN.Furthermore, modified RBFNN is adopted to building converter distillingvanadium static mathematical model to off-line control production process. Compared with traditional RBFNN control result, it is shown that, after training RBFNN hidden-output nodes weight using chaos genetic algorithm other than using grads decline method to rectify nodes weight., model is easier to reach entire optimization and enhance precision and rapidity of approximating non-linear system. Moreover converter distilling vanadium model based on modified RBFNN has the property of self-learning, self-organization and self-adaptive. After using the model to predict assistant materials quantity, the hitting target of every desire of vanadium, carbon and middle-stage-steel temperature is enhanced. So the model has practicable value.
Keywords/Search Tags:Radial Basis Function Neural Network(RBFNN), chaos genetic algorithm, crossover and mutation probability, approximation, converter distilling vanadium static model
PDF Full Text Request
Related items