Hot metal desulphurization control model based on PCA-RBFNN

Hot metal desulphurization pretreatment process has characteristics of multivariate and nonlinear, the sulfur content of hot metal can not be monitored online, it´s difficult to set the amount of desulfurating agent accurately. To solve this problem, a model of RBF neural network was proposed based on production data of desulphurization. The model improved quality of the modeling data by removing outliers with similarity coefficient method. Simplified the model structure and reduced data noise by eliminating correlation of the modeling data with PCA method. Improved generalization ability with subtractive clustering algorithm and improved network error learning function. Simulation results show that the model is high accuracy and can determine the amount of desulfurating agent accurately.