Abstract: A principal component fuzzy neural network model based on kernel method and greedy algorithm is proposed,in order to improve the carbon contents of furnace endpoint and precision of the temperature focasting model.The model adopts kernel function to project the input variables into high dimensional feature space,so that the latent information can be extracted.Then greedy algorithm is used to select principal components,remove redundant information and reduce the input dimension.After the extracted principal components are introduced into the adaptive neuro-fuzzy inference system(ANFIS),the network reveals the implication relations among the inputs by means of rules,so as to simulate experience of the operators and consequently to reduce the in?uence resulted from different operators.Simulations are made with practical data,and the result proves the validity of the proposed model.