Research And Application Of Blast Furnace Molten Iron Quality Prediction Method Based On Kernel Extreme Learning Machine

As the main industry of the entire national economy,the steel industry has played an important role in promoting the rapid development of the national economy.Blast furnace ironmaking is an important part of steelmaking production.As the internal reaction of blast furnace is extremely complicated.The solid,liquid and gas are multi-phase coupled,and have complex dynamic characteristics such as strong coupling and large time lag.Therefore,modeling research on it is very difficult.With the development of industry,domestic and foreign scholars pay more and more attention to the improvement of blast furnace ironmaking process and the research of ironmaking principle.The modeling of blast furnace ironmaking process and the control research of the whole ironmaking system had always been the key research problems in the field of metallurgical engineering and automatic control today.At present,most of the modeling methods for molten iron quality(MIQ for short)parameters are to model a single molten iron content or molten iron temperature parameter.Since a single parameter can singly reflect a certain state of the blast furnace reaction process,it cannot reflect the comprehensive state of blast furnace production.It cannot meet the operational and control requirements of actual blast furnace ironmaking production.In addition,the traditional neural network method for modeling the quality parameters of molten iron has problems such as slow convergence rate,easy to fall into local optimum,structure parameters of model cannot be optimal,and the model prediction accuracy is not enough,and the generalization ability is poor.Therefore,in view of the above problems,this paper proposes a research on the modeling method of molten iron quality parameters based on kernel extreme learning machine.The specific work is as follows:(1)Analyzing the principle of blast furnace ironmaking,studying the dynamic characteristics of blast furnace ironmaking process,determining the modeling parameters of blast furnace molten iron quality,then briefly describe the blast furnace data collection process,and preprocess the data of a large number of blast furnace body data obtained.It includes missing value processing of sample data,outlier processing,smoothing of data,normalization processing,and the like.(2)Aiming at the extreme learning machine,it is a single hidden layer.Each training randomly initializes the input weight and the hidden layer offset vector,which leads to the instability of the model structure.A prediction model based on the kernel extreme learning machine’s molten iron quality parameters is proposed.The kernel function is introduced to replace the feature map of the hidden layer nodes of the extreme learning machine,so that the number of nodes of the hidden layer is not required to be artificially set,and the training time of the model is reduced.In addition,in order to reduce the computational complexity of the model,the principal component analysis method is introduced to reduce the dimensionality of the input variables affecting the quality parameters of the molten iron,which effectively improves the accuracy and efficiency of the modeling.At the same time,the kernel extreme learning machine is compared with the traditional BP neural network,SVM and ELM.The simulation results show that the kernel extreme learning machine model has higher prediction accuracy and the generalization performance.(3)The introduction of kernel parameters for the kernel extreme learning machine results in very sensitive changes to the structural parameters of the model,and the structural parameters of the model are not optimal.Therefore,combining with differential evolution algorithm(DE for short)and the particle swarm optimization algorithm(PSO for short),and borrowing the meme evolution mechanism of shuffled frog leaping algorithm(SFLA for short),a new hybrid intelligent optimization algorithm(DEPSO for short)is proposed.The DEPSO is used to optimize the structural parameters of the kernel extreme learning machine model to solve the optimal structural parameters.Finally,the DEPSO-KELM molten iron quality parameter prediction model is established,and then a lot of historical data collected from the blast furnace ironmaking field were used for model training and testing.The simulation experiments show that the accuracy and generalization performance of the DEPEO-KELM molten iron quality prediction model is obviously improved,which meets the requirements of the actual industrial site.(4)Taking the blast furnace ironmaking process of a steel plant as the research object,based on Movicon X software,the blast furnace molten iron quality monitoring system was designed and developed.Real-time monitored the important process parameters and molten iron quality parameters of blast furnace ironmaking process.Analyzing and diagnosing the quality of molten iron,then providing early warning information and guidancing to the blast furnace operators,adjusting the furnace condition of the blast furnace in time to ensure the smooth operation of the blast furnace and produce qualified molten iron.