Prediction Of Silicon Content In Hot Metal Of Blast Furnace Based On Nonlinear Combination Model

The iron and steel metallurgy industry has always been one of the pillar industries of China's national economy and has a decisive role in the development prospects of China's economy.Blast furnace ironmaking,as an important process that is a major component of the steel manufacturing industry,has always been regarded as an indispensable and critical component of the steel industry.The stability of the furnace state in the smelting process is closely related to whether the process can be carried out in an efficient and energy-saving manner.In the process of blast furnace smelting,furnace temperature is often used as a monitoring indicator for real-time control to ensure the smooth running of the smelting process.The silicon content has long been chosen as an important indicator of the thermal condition of the blast furnace.The design of a stable and accurate silicon content prediction model to guide the blast furnace operator in the furnace temperature regulation is the significance of this thesis.The study of silicon content prediction models is a critical guide for the production practice of the iron and steel industry,while providing an in-depth investigation of the underlying theory.Both data and practical results show that many uncertainties can lead to variations in furnace temperature.The data show that such industrial indicators,which are prone to fluctuations in silicon content,usually have significant nonlinear characteristics.Considering that least squares support vector machine has good learning ability and low generalization error rate,and it usually performs well in the face of data with strong nonlinear characteristics,this thesis chooses to use it as the main body to build a nonlinear combined prediction model to predict the online production data of Baosteel Group.Considering that artificial neural networks have good performance in nonlinear mapping and can be self-optimized with high error tolerance,BP neural network,which is the most widely used in the field of time series data,and Elman neural network,which improves stability by adding implicit layers,are selected as the combined objects in this thesis.The combined results are used as the training set of the LS-SVM nonlinear combination model,and the actual values of the corresponding ovens are selected as the test set.The LS-SVM model is then used to make predictions again,and the corresponding predicted values of silicon content are obtained.The simulation results show that the mean relative error of the nonlinear combination model is 0.022,which is 65.6% and 35.3% lower than that of the BP and Elman neural networks respectively;the mean absolute error is 0.019,which is 60.4% and 38.7% lower than that of the two models respectively;the root mean square error is 0.00053,which is85.3% and 64.4% lower than that of the two models respectively.and 64.4% respectively.The significant reduction in all error values as a measure suggests that the non-linear combination model is more accurate,better captures the composite characteristics of the time series data,thereby capturing more dynamic information,and effectively addresses the prediction bias of the single model.Therefore,compared with a single forecasting model,the nonlinear combination model has better forecasting ability and more accurate forecasting results.