Bilinear Subspace Modeling And Nonlinear Predictive Control Of Molten Iron Quality Indices In Blast Furnace

For a long time,the steel industry has played an irreplaceable role in the national economy,providing an important raw material guarantee for national construction.As one of the most important links in the steel production process,the blast furnace(BF)ironmaking will bring great economic and social benefits owing to each step of technological progress.The objective of BF ironmaking is to realize high production and low energy consumption.In order to achieve this goal,it is necessary to provide effective control and real-time monitoring for the internal smelting condition of BF,especially the molten iron quality(MIQ)indices for ironmaking products.At present,four indices,namely the Si content([Si]),the phosphorus content([P]),the sulfur content([S])and the molten iron temperature(MIT)are widely used to characterize the MIQ.Among them,[Si]and MIT are the two most concerned indices for the operators of BF.However,the MIQ indices can only be detected by off-line test,which usually takes 2?3 hours to get the detection value of MIQ.Meanwhile,BF ironmaking is a complex nonlinear process,which the internal environment of the BF is extremely complex,often accompanied by high temperature,high pressure and the coexistence of solid,liquid and vapor multiphase,making it difficult to establish a mechanism model of MIQ.At present,the control of MIQ often depends on the operator's experience and judgment,which lacks scientific guidance.Therefore,in order to establish an effective model and realize the control and optimization of MIQ indices for its stable running,it is necessary to establish a dynamic model and provide effective control of MIQ indices for BF ironmaking process by means of data-driven modeling method.Considering above problems,this paper relies on the major projects of the National Natural Science Foundation,which includes two parts,“High-performance operation control and implement technology of large blast furnace"(61290323)and"Experimental verification platform construction and application verification of large blast furnace high-performance operation control"(61290321),respectively.Moreover,the paper carries out the research on bilinear subspace modeling and nonlinear predictive control of multivariate MIQ indices in blast furnace ironmaking and conduct experiment on 2#blast furnace of Liuzhou Iron and Steel Company in Guangxi province.The specific works are as follows:(1)Firstly,starting from the whole process of BF ironmaking and considering the controllability and observability of variables,a filter-Wrapper feature selection method based on machine learning is used to solve the problem that the large number of variables and the coupling correlation in ironmaking process make it hard to establish the model of MIQ indices.For Filter,each feature is scored according to the divergence or correlation of variables,and the feature is selected by setting the threshold or the number of feature selections.For Wrapper,according to the set of objective function,several trainings are carried out,which some features are selected or excluded.The final selected controllable and measurable variables as the modeling inputs of control-oriented MIQ,namely flow rate of cold air,pressure drop,flow rate of rich oxygen and volume of coal injection.(2)Secondly,considering the difficulty of establishing the mechanism model and the shortcomings of existing linear modeling methods in predictive control of MIQ indices,a subspace identification method based on bilinear system model is proposed for predictive control of MIQ indices.First,on the basis of linear subspace model,a bilinear subspace model is introduced to compensate the nonlinear dynamic characteristics of BF ironmaking process and improve the accuracy of modeling.Then oblique projection and SVD matrix decomposition are used to reconstruct the state matrix,and least square method is used to extract the system matrix.Finally,based on the acquisition of input and output data of BF ironmaking process,a data-driven bilinear subspace model for the control of MIQ indices is established.The simulation experiment is carried out by using actual BF data,and the results show that,compared with the conventional predictive control of MIQ indices based on linear subspace identification,the proposed method has higher model accuracy,better setpoint tracking performance and stronger interference suppression ability.(3)Finally,aiming at the problem that that the complex nonlinear dynamic characteristics of BF ironmaking process are difficult to control effectively,a data-driven recursive bilinear subspace for online predictive control of MIQ indices is proposed.The proposed method is used to establish a data-driven online model for predictive control of MIQ indices by using recursive bilinear subspace identification with forgetting factor.Unlike conventional linear subspace and offline bilinear subspace modeling,this method constructs extended Hankel matrix to get subspace matrix,which is used to construct the data-driven input-output model without explicitly estimating the system matrix can effectively reduce the computational complexity.The established prediction model is applied to the predictive controller design of MIQ.Since the subspace matrix of the predictive model is updated online by the latest process data,the adaptive predictive control of MIQ indices can be realized.The simulation experiment based on the actual data of BF shows that the proposed method can effectively reduce the fluctuation of BF ironmaking process,and has more stable and reliable control performance.