Research On Quality Modeling Of Blast Furnace Ironmaking Processes Based On Improved Incremental Random Vector Functional-link Networks

Blast furnace ironmaking has always been the most important way of ironmaking in the world,accounting for more than 90%of the total ironmaking.The blast furnace smelting process is continuously carried out.Only by ensuring the stable material flow movement and the stable energy supply-demand relationship in the furnace,can high-yield,high-quality,low consumption and low cost be achieved,so as to achieve good production technical indicators and economic benefits.However,the internal reaction of the blast furnace is complex,high temperature,high pressure and coexistence of gas-liquid-solid multi-state,the smelting environment is bad,which makes it difficult for field operators to monitor and control the internal state in real time.In the actual blast furnace ironmaking process,the molten iron quality parameters(molten iron temperature(MIT),silicon content([Si]),phosphorus content([P])and sulfur content([S]))are usually used to indirectly reflect the internal state of the blast furnace.However,the existing detection technology is difficult to measure on-line,and the off-line test has a long lag time,so it can not be fedback in time,which seriously affects the optimal operation of blast furnace.Therefore,it is very necessary to establish an accurate and reliable prediction model of molten iron quality for the stable and smooth operation of blast furnace.In recent years,random vector functional-link neural networks(RVFLNs)have been widely used in industrial process quality index modeling because of its high efficiency.However,the existing random vector functional-link neural networks,such as basic random vector functional-link neural networks,usually have the problems of unstable calculation results,low prediction accuracy,prone to overfitting and difficult to select the optimal number of hidden layer nodes.The incremental random vector functional-link neural networks solves the problems of the basic random vector functional-link neural networks in the modeling process,but the convergence speed is slow,the operation efficiency is low,and the final structure of the model is complex.Aiming at the above problems,this paper relies on the National Natural Science Foundation’s major project "High-performance operation control method and implement technology of large blast furnace"(Project No.:61290323)to carry out the research on the quality modeling of blast furnace ironmaking process based on incremental random vector functional-link neural networks.The specific research work is as follows:(1)Aiming at the problem that the network parameters are difficult to be optimally determined and the model convergence speed is slow in the traditional incremental random vector functional-link networks(I-RVFLNs),an optimized incremental random vector functional-link networks(O-I-RVFLNs)modeling algorithm for molten iron quality is proposed.Different from the traditional I-RVFLNs,the proposed O-I-RVFLNs algorithm first sets a desired modeling residual error vector,and then selects the input weights and biases that can reach or less than the expected residual error as the input parameters of the node each time a hidden node is added,so as to improve the convergence speed of the model.In addition,in consideration of the problem that the modeling error is getting smaller and smaller and the decreasing trend is becoming less obvious during the continuous iterative update process,the RMSE difference between adjacent iterations of each index parameter is considered in the termination condition of the algorithm,and the corresponding convergence criteria are formulated by referring to the Western Electricity Rules in statistical process control.Finally,based on the standard data set and actual blast furnace industrial data,the proposed O-I-RVFLNs algorithm was verified and applied.The results show that compared with other RVFLNs algorithms,the data model established by the proposed algorithm has faster convergence speed and better generalization performance.(2)The output weights obtained by the traditional I-RVFLNs model are not the least squares solutions,and after the model training,there are too many hidden nodes in the network,which leads to the model structure is too complex.Aiming at the above problems,an improved orthogonal incremental random vector functional-link neural networks(I-OI-RVFLNs)algorithm is proposed,and a blast furnace molten iron quality prediction model based on the I-OI-RVFLNs algorithm is established.First,the Schmitt orthogonalization method is used to orthogonally calculate the output vector of each hidden layer node in the network,and then the least square solution of the network output weight is obtained,so the model convergence speed is improved.In addition,in order to reduce the number of hidden nodes in the network after model training and simplify the model structure,the proposed algorithm further changes the network construction mode of the existing I-RVFLNs,which is to approach the network output by fixing the number of hidden nodes in advance and then iteratively modifying the output weight of each hidden node.Finally,based on the actual blast furnace data,a multi-element prediction model of molten iron quality based on I-OI-RVFLNs is established and compared with the modeling results of other incremental random vector functional-link neural networks.The results show that compared with the traditional I-RVFLNs,the existing enhanced incremental random vector functional-link neural networks(EI-RVFLNs)and the orthogonal incremental random vector functional-link neural networks(OI-RVFLNs)models,the model established by the proposed algorithm can achieve faster convergence speed and more compact network structure.