| The iron and steel industry provides an important raw material guarantees for national development and is an significant basic industry of national economy.As a primary steelmaking method in China,BOF(basic oxygen furnace)steelmaking owns high productivity and relatively low production cost.BOF steelmaking is a complicated and high temperature physical chemical process in which raw materials,such as molten iron,steel scrap and pig iron,are converted into steel.As the key production indicators,the endpoint carbon content and temperature of the molten steel determine whether or not the quality of steel meets the standard.Accurate prediction of the endpoint carbon content and temperature of the molten steel can reduce the number of supplementary blows,which is of great significance for improving production efficiency and energy saving and emission reduction.Due to the limitations of on-site environmental conditions and technology,it is difficult to accurately measure the carbon content and temperature in real time.In the process of BOF,the steelmaking process data such as the amount of raw materials added,the amount of oxygen blowing,and the oxygen blowing time are closely related to the end carbon content and temperature.Therefore,the soft sensor models are built to predict the end carbon content and temperature in this paper.However,with the development of steelmaking technology and the increase in steelmaking equipment,the number of attributes/features of BOF steelmaking data has become more enormous than before.Additionally,the quality of raw materials from different batches varies greatly,and the law of steelmaking process samples has changed.The imbalance problem frequently arises in the practical production data.This problem also arises in the BOF steelmaking process data.These problems make it very challenging to accurately predict the endpoint carbon and temperature of the molten steel.To deal with the problems of high dimensionality,time-varying as well as imbalance in in production process data of BOF steelmaking,based on a data-driven method,this paper preprocesses the BOF steelmaking process data by selecting the significant features,and uses just-in-time-learning strategy to measure similarity between samples with time-varying characteristics.And then local soft sensor model is built to predict the endpoint carbon content and temperature.The details of main research are described as follows:(1)Feature selection of BOF steelmaking process data based on improved salp swarm algorithm.In order to eliminate irrelevant or redundant data and improve the accuracy of basic oxygen furnace(BOF)steelmaking endpoint prediction model,a novel denary version of the salp swarm algorithm(SSA)is proposed in this paper and applied for feature selection of BOF steelmaking process data in wrapper mode.Firstly,the proposed denary SSA presets the dimension of solutions instead of the strategy of indeterminate number that will lead to different results over various runs.Then the native and binary versions of SSA are applied to generate candidates for leader salp;meanwhile,a probability function is utilized in DSSA to replace each element of leader salp.Finally,an update strategy for follower salps is used to enhance the exploitation of the SSA algorithm.The proposed method is employed to find the optimal solution that maximizes the regression accuracy and minimizes the non-repeatability of the feature selection on BOF steelmaking process data.The performance of the proposed approach is compared with various state-of-the-art approaches in terms of different assessment criteria.Results show that the proposed denary SSA approach of feature selection provides the repeatable results and obtains higher regression accuracy.The proposed method is used to reduce the fitness function value to the optimal value,and the optimal feature subset is selected from the BOF steelmaking process data.Irrelevant and redundant features can be eliminated using the proposed algorithm,which is very necessary for the prediction of end point carbon and temperature in molten steel.Under the circumstance that the tolerable error of carbon content is ±0.02%,the accuracy is88.00%.Under the circumstance that the tolerable error of temperature is ±10℃,the accuracy is 83.33%.(2)Just-in-time-learning based prediction model of BOF endpoint carbon content and temperature via vMF mixture model and weighted extreme learning machine.In BOF steelmaking,the quality of raw materials varies greatly between different batches,which would lead to the inaccurate predictions for these two indicators.Additionally,there are imbalance problems in production process data of BOF steelmaking.For the first problem,a novel similarity criterion based on von-Mises Fisher mixture model(VMM)is proposed in this paper and applied for sample selection of just-in-timelearning(JITL)-based endpoint carbon content and temperature prediction model.The V-shaped transfer function is utilized to develop weighted extreme learning machine(WELM)as local regression model to address the imbalance problems.The performance of the proposed methods is compared with other methods under JITL framework.Using the proposed similarity criterion based on VMM and the WELM,the JITL-based endpoint carbon content and temperature soft sensor model can be bulit,which can avoid the problem of concept drift and realize the online soft sensor of the endpoint carbon content and temperature.Under the circumstance that the tolerable error of carbon content is ±0.02%,the accuracy of this paper is 93.98%.Under the circumstance that the tolerable error of temperature is ±10℃,the accuracy is 92.48%.(3)Just-in-time-learning based prediction model of BOF endpoint carbon content and temperature via optimized moving windows.In BOF steelmaking production process,as time goes by,there will be equipment aging,the quality of raw material changes,and production requirements changes.This will lead to large differences in samples at different periods of time.It is difficult for a single prediction model to apply to a new query data,which will reduce the prediction accuracy.For the time-varying problem,a novel just-in-time-learning strategy based on optimized moving windows is proposed in this paper.Firstly,the relevant samples are selected from database based on similarity criterion.Take similar samples as the reference point,use the moving window to find the appropriate time nearby samples.Then noisy data are removed from the windows by optimization algorithm.Finally,the time nearby samples in the optimized moving windows are used as the training set,and a local soft-sensing model is established to predict the end point carbon content and temperature.When the time span of the BOF steelmaking data is large,the time-varying problem can be effectively avoided,and the end point carbon temperature can be accurately predicted by the proposed method.This method has certain practical application value for the prediction of the BOF steelmaking endpoint carbon content and temperature.Under the circumstance that the tolerable error of carbon content is ±0.02%,the accuracy of this paper is 90.50%.Under the circumstance that the tolerable error of temperature is±10℃,the accuracy is 93.17%. |
PDF Full Text Request