Research On Prediction Of Blast Furnace Tenperature Based On Fuzzy Distributed Model

The blast furnace iron making process is a highly complex nonlinear process, in the course of itsoperation, the furnace conditions continue to change over time. Temperature control is the keyto efficient production and to stabilize the blast furnace. Blast furnace production process haslarge number of data, noise, highly coupled, and data conflicts, that result in a single model isdifficult to establish accurate blast furnace temperature prediction model. In the context of theBlast Furnace smelting process characteristics, Issues, based on data-driven modelingtechniques, combining the fuzzy theory, the idea of distributed modeling, establish fuzzydistributed neural network blast furnace temperature prediction model. The model establishessub-models for the blast furnace, and then combines the sub-models. This model improves theprecision of the prediction model. Specific study content is as follows:1. The data collected in the field from blast furnace have missing values and outliers. Forthe blast furnace is a large time delay system, that many factors affecting the furnacetemperature have a time lag. In order to establish a more accurate model, preprocess site data,take smooth handling of missing values and outliers, use correlation analysis to calculatecorrelation coefficient of the process parameters. Select the model parameters that having agreater impact on the furnace temperature as the input variables. Analyze the lag time ofinfluence on output variables in the blast furnace production process.2. Because the hot metal silicon content and the hot metal temperature can both reflect thechange of blast furnace, the hot metal silicon content and temperature of molten iron are as theoutput variable respectively, and the selected process parameters are as input variables toestablish fuzzy distributed neural network model. Use the improved fuzzy clustering algorithmto fuzzy partition input and output sample data into several subspaces. Then establish the subnetneural network by the several subspaces. Finally combine with the subnets by membershipdegree to establish the distributed neural network and fuzzy blast furnace prediction model.Combining with the practical situation of blast furnace and comparing the predicted effect of hot metal silicon content and temperature of molten iron, choose the better predicted effect ofhot metal temperature as output variables.3. The fuzzy clustering algorithm and neural network modeling method in distributedneural network model can have a variety of different methods, so select fuzzy c-meansalgorithm and the self-organizing neural network clustering algorithm as clustering algorithms,and choose BP neural network and RBF neural network as the sub-modeling. Combining ofboth two, four different fuzzy distributed neural network models are established. Comparedwith the four different models, distributed BP neural network model based on FCM is better.4. In order to get a better predicted effect, using particle swarm optimization algorithm tooptimize distributed BP neural network model based on FCM to determine the subnet neuralnetwork weights and thresholds of the model. Through the accuracy validation, the predictedeffect of the model has been improved.Through testing the established prediction model by the collection of data, the results showthat distributed neural network model can achieve expected effect better. It has a certain guidingsignificance of blast furnace operation.