Experience in implementing machine learning to calculate the quality and production of agglomerate

Abstract

In 2020, EVRAZ United West Siberian Metallurgical Combine JSC (EVRAZ ZSMK JSC) completed work on the creation of a mathematical modeling system for all processing units of the metallurgical plant. During testing of the system, a high error was found in the existing factor model for predicting agglomerate production, which was developed taking into account the specific sintering rate of individual concentrates. The paper proposes the use of linear regression to predict the productivity of sintering machines, which, unlike nonlinear methods, is optimal for integration into high-performance optimization systems. A feature of the work is forecasting, taking into account the proportion of the agglomeration charge. The model was tested at EVRAZ ZSMK JSC and showed sufficient accuracy (R2 > 90). A large economic effect is expected from the model. A separate study of existing agglomerate quality forecasting systems was conducted. Machine learning (ML) methods have recently made a great contribution to the development of forecasting models used to assess the quality of the agglomerate. This is due to the fact that the sintering process is a very complex dynamic with non‒linearity and a large delay. The physico-chemical phenomena involved in this process are complex and numerous. The neural network can constantly adjust the parameters of the model to reflect changes in systemic causes. A linear method was also studied to predict the agglomerate quality. Due to the poor quality of the resulting linear model, the “random forest” machine learning method was applied. Currently, the model is being operated as part of the integrated optimization program SMM Prognoz for the entire plant. For the convenience of the user, when implementing the module, visualization of the model quality using historical data was added, as well as the statistical metrics obtained.