System analysis of the factors determining the stability of the converter lining and the predictive model construction

Abstract

In the course of this work, the process of wear of the converter lining was considered. The latter can withstand thousands of smeltings, each of which, in one way or another, affects the service life of the lining. The experience of 91 sessions of lining work for seven years was studied (in total - more than 145 thousand smeltings). The purpose of the study is to develop a family of models of dependencies of the stability of the converter lining from known technical and technological factors to further provide recommendations for the production process.
From a wide range of factors recorded on the selected object of study, 15 statistically significant factors were identified, and 13 of them were numerical, and 2 were categorical. Two of the numerical factors have a strong correlation and cannot be used in forecasting models at the same time. In the analysis of experimental data, three of the 91 studies were rejected because they contain emissions. The rest of the data using the Ward’s method are divided into two clusters, in which significantly different experimental results are identified.
For each of the clusters, a statically significant regression model for predicting the stability of the lining based on the internal criterion of minimum complexity and the external criterion of regularity was obtained. Also, using the method of least squares based on internal and external criteria, a universal model of multiple linear regression is obtained, which describes the dependence of the stability of the lining on six significant factors with a coefficient R2 greater than 0.9. The remnants of the model have a normal distribution, and their variance is relatively constant.
Alternatively, a model using a self-organizing reference function with allowable nega-tive degrees of arguments is proposed. The Akaike criterion was used as an internal criterion for the selection of partial models, and the external criterion of standard cross-validation was used to select the best models. The resulting alternative model has 77 real coefficients and is based on five factors.
Due to the high estimates of all obtained models according to statistical criteria in the future, it is recommended to use the multiple linear regression model to estimate the stability of the lining. At the same time, the complex model obtained by GMDH can be used for a more subtle analysis of the influence of individual factors on the established mode of operation of the lining.
It is proposed to use the obtained models as part of an integrated management system of a metallurgical enterprise, namely in the decision support subsystem of the converter operator to form advice on increasing the duration of the current lining.