Research On Endpoint Prediction Model Of Converter Steelmaking Based On Support Vector Regression Machine

The demand for steel continues to increase with the process of social industrialization,and the requirements for quality are also continuously improved.The production of high-quality steel is inseparable from the steelmaking end point control system.The core issue of endpoint control is to establish an endpoint prediction model.At the same time,the carbon content and temperature in molten steel at the end of steelmaking are the key factors that determine whether the quality of the tapped steel meets the requirements of the type of steel being made.Therefore,it is of great significance to establish an accurate prediction model for the end point of converter steelmaking for steel smelting.Compared with other theoretical methods,the SVR algorithm has good explanatory properties,strict mathematical derivation,and solves the problem of easy falling into local optimum,so it is suitable for industrial production.In view of the problems existing in the application of existing models in the prediction of the end point of converter steelmaking,this thesis has done research on the following aspects after understanding the principle of converter steelmaking production:(1)The predictive model is built using historical melting data in converter steelmaking.The data set will inevitably affect the effect of the model,so the original data needs to be processed.First of all,since the smelting process of the same steel type is relatively fixed,for the problem of missing some values in the data set,the K nearest neighbor interpolation(KNNI)method is used to find k similar samples adjacent to it,and assign different Gaussian weights to them.Weight,the missing data is filled by the weighted average of these k samples.Secondly,the partial correlation coefficient method and the gray relational analysis(GRA)method are combined with the entropy weight method,and the degree of correlation between input features and between input features and outputs is calculated.The selection of input features is completed,which effectively prevents the difficulty of modeling caused by feature redundancy.(2)In order to reduce the impact of singular value samples and outlier samples on the model endpoint hit rate,on the basis of SVR and LDMR related theories,a ε margin optimization based weighted support vector regression(ε-MOWSVR)was developed propose.In order to avoid the prediction results of the model being affected by singular value samples that fall near the ε interval,the loss of samples inside and outside the interval is calculated at the same time.The sample ensemble distribution is optimized by trade-off minimizing the ε interval loss and the sample ensemble loss.In addition,by introducing density weight information,the impact of outlier samples on the results is reduced.In addition to conducting experiments on the UCI dataset and the original dataset of converter steelmaking,the effectiveness of the improved strategy is further verified by artificially adding singular value samples and outlier samples to the dataset.(3)In order to solve the problem of low parameter search efficiency of ε-MOWSVR,this thesis proposes the Predator Competitive Optimization Algorithm(PCOA)based on swarm intelligence,and combines it with ε-MOWSVR to predict the end point of converter steelmaking.The ε-MOWSVR-PCOA algorithm is established.The algorithm optimizes parameters through a random search strategy,and the problem of local optimum is solved.Experimental results show that this method can achieve a higher endpoint hit rate while saving parameter search time.