Data-Driven Robust LS-SVR Modeling And Nonlinear Predictive Control Of Blast Furnace Ironmaking Process

The iron and steel industry,as one of the most important pillar industries in China,plays an irreplaceable role in the development of national economy and society.Its theoretical research and industrial applications has always been a hot topic in many fields of metallurgy,automatic control and technology etc.In order to realize high production,low consumption and stable operation of blast furnace(BF)ironmaking process,the real-time monitoring and effective control of BF ironmaking process are needed.Due to the limitation of current detection methods,the molten iron quality(MIQ)parameters such as silicon content([Si]),phosphorus content([P]),sulfur content([S])and molten iron temperature(MIT)are difficult to be detected online directly,and time lags of the off-line test is longer.Meanwhile,owing to the complex physical and chemical reactions,heat and mass transfer,and multi-phase flow'under the conditions of high temperature,high pressure,multi-phase and multi-field coupling,it is difficult to establish the mechanism model of MIQ under the complex characteristics of BF ironmaking process.Therefore,the MIQ model and control on the basis of above model for BF ironmaking process must be established by using artificial intelligence and statistical learning techniques to solve the problems of unstable production and low product quality caused by the current monitoring and control of BF ironmaking process which relies heavily on operators' experience and judgment.Considering above problems,this paper carries out the research and application on data-driven robust LS-SVR modeling and nonlinear predictive control for BF ironmaking process under the framework of statistical learning theory and conduct experiment on#2 BF of Liuzhou Iron and Steel Company in Guangxi province with the support of National Natural Science Foundations,Specific works are as follows:1.To solve the problem that the parameters of silicon content([Si])of MIQ was difficult to be detected directly and be obtained by manual analysis,which brought large time delay,a method of sparse and robust least squares support vector regression(R-S-LS-SVR)was proposed to establish the dynamic modeling of[Si]with the help of multi-objective genetic optimization of model parameters.First,PCA is applied to deal with numerous variables and multivariate coupling.Then the maximal independent set of sample data in feature space mapping set was extracted to realize the sparse of training sample set and reduce the computational complexity of modeling.And a method to improve the modeling robustness,which obtains a robust R-S-LS-SVR model,was proposed by introducing IGGIII weighting function into the obtained S-LS-SVR model.Meanwhile,the multi-objective evaluation index that synthesizes the modeling residue and the estimated trend was presented to compensate the single RMSE index.Based on it,an on-line soft sensor model of the hot metal[Si]with the optimal parameters was obtained by using the multi-objective genetic algorithm(NSGA-II)with the non-dominated sort and elitist strategy.Finally,the R-S-LS-SVR was applied to the prediction of molten iron[Si]in BF ironmaking process.The simulation verification and analysis showed that sparseness and robustness of the improved algorithm are effective and advanced.2.A robust multi output least squares support vector machine(R-M-LS-S VR)algorithm based on transfer learning and the M-estimation was proposed to solve the modeling problem of MIQ parameters.First,to completely capture the nonlinear dynamics of the BF process,the nonlinear autoregressive exogenous(NARX)model was constructed for the MIQ indices.Meanwhile,considering the standard LS-SVR cannot directly cope with the multi-output problem,the multi-task transfer learning was proposed to design a novel multi-output LS-SVR(M-LS-SVR)for learning of the NARX model.Furthermore,a novel M-estimator was proposed to reduce the interference of outliers and to improve the robustness of the M-LS-SVR model.Finally,a novel multi-objective evaluation index on modeling performance was developed by comprehensively considering the root mean square error of modeling and the correlation coefficient on trend fitting,based on which the NSGA-II algorithm was used to globally optimize the model parameters.According to actual industrial data and data test illustration,the proposed method could efficiently eliminate the adverse effect caused by the data fluctuation in BF process with stronger robustness and higher accuracy.Moreover,control test showed that the developed model can be well applied to realize data-driven control of the BF process.3.In regard to the difficulty in effectively control the MIQ index by conventional methods,a data-driven nonlinear predictive control method was further studied on the basis of a robust modeling above.First,a nonlinear predictive control based on R-S-LS-SVR prediction model was proposed to solve the control problem of[Si]for BF ironmaking process,and the dynamic optimization was carried out by using the genetic algorithm(GA)with global optimization in the real-time optimization process.Then,aiming at the control problem of[Si],[S],[P],MIT etc.,a data-driven nonlinear predictive control method for the multiple MIQ index based on multivariable inverse system decoupling was proposed by using the proposed R-M-LS-SVR to establish the prediction model and inverse system model of the process.Finally,the actual industrial data was used to verify the effectiveness and advancement of the proposed control method,so as to provide a solution to the closed-loop control of the whole process of the BF ironmaking process.