Prediction Method Of Silicon Content In Blast Furnace Hot Metal Based On Improved BP Neural Network

In order to achieve the comprehensive goal of high efficiency,low energy consumption,long life and low pollution,modern iron smelting technology has gradually become large and efficient in production mode.The blast furnace is a nonlinear complex system with large time delay,and the high temperature and high pressure environment make it difficult to measure and control the temperature.In this paper,the indirect prediction of furnace temperature is achieved by predicting the silicon content of molten iron by using the positive correlation between the silicon content of molten iron and the furnace temperature of the blast furnace.In order to solve the problem that there are many influence factors in the prediction of silicon content,this paper uses the ability of Neural Network to realize complex nonlinear mapping to build BP Neural Network model to predict the silicon content of hot metal in blast furnace.In this paper,a gray relational analysis is made on the actual production data of a steel blast furnace.Ten factors that are highly correlated with the silicon content of molten iron,such as blast air humidity,furnace top temperature and furnace top pressure,are selected as the input of the prediction model.Before establishing the model,the dimensionality of each data was unified through normalization processing,and the lag time of each factor was studied and determined.In view of the problems of unstable prediction performance and large sample error of the BP neural network prediction model established in the early stage,this paper puts forward a scheme to improve the traditional prediction model with the advantages of strong global search ability and high fault tolerance of Genetic Algorithm and establishes GA-BP model simulation experiment.The simulation results show that the optimized prediction model has certain improvement in prediction stability and hit ratio.Finally,the APSO-BP model is established and simulated by the Particle Swarm Optimization model with the advantages of fast convergence and high efficiency.By comparing the results of the three prediction models,it is found that the model optimized by the improved Particle Swarm Optimization algorithm has the best performance and can achieve the prediction within 4.8% of the test sample error,which proves the feasibility of the prediction method.