Some Application Of Multi-model Fusion Technology For Prediction Of The EAF Endpoint Temperature

Complex industrial systems have the charaeteristies of high dimension of inputs, nonlinearity, strong coupling, wide range of conditions, integrated high performance requirements, etc. Using intelligent methods such as neural network, fuzzy logic to establish a single model, often difficult to fully describe the global properties of complex systems. In this case, multiple model approach, which based on divide-and-conquer principle is an effective solution, the basic idea is to complex large problem into a simple little question, then solve the small problems of integration by some way and get the original problem solution. Because of this method has the ability to improve model accuracy and generalization characteristics is applied to the modeling of complex industrial systems.This paper presents a fuzzy clustering method based on multi-model fusion modeling. Using this fuzzy c-means clustering (FCM) algorithm, a system could be quickly divided into multiple optimal fuzzy parts, the cluster number corresponded to the optimal number of sub_systems, and using BP neural network to establish sub-model, multiple sub-model uses a weighted approach to integration, at this, determination of weights is the key issue of the multi-model fusion modeling. This paper presents Bayesian fusion method, weights to determine the multi-model fusion modeling of the key issues determination of weights is the key issue of the multi-model fusion modeling, as priori probability for the Bayesian algorithm is difficult to determine, using the sample set based on clustering as a priori knowledge. Then estimated prior probabilities and class conditional probability sample points based on the class of data set, thus obtained sample points using Bayesian formula the posterior probability, using the posterior probability as a child model since the weights, weighted integration, to get the final output of the model.EAF steel-making process is a very complex physical and chemical processes, which bath temperature is too high, smelting conditions are too harsh, continuous bath temperature can not be directly detected. Establish an appropriate model to predict the endpoint temperature of EAF. Will play an important role in the temperature control. A fuzzy clustering method based on multi-model fusion modeling is applied to the endpoint temperature of EAF. Simulation results show that the multi-model fusion modeling has higher predictive precision and hitting accuracies, compared to BP single model.