Abstract
A new approach is proposed to predict the silicon content in hot metal with Bayesian networks. Some key variables, affecting hot metal silicon content, were selected out and analyzed. Then a Bayesian network (BN) model was constructed according to the causal relationship of those variables. And the parameters of the model were estimated with the data selected from No.1 BF in Laiwu Iron and Steel Group Co.. Finally an improvement was made on BN method by defuzzification methods. The results show that the prediction is very successful and Bayesian network is better than BP neural network due to the visible inference and convictive results.