Intelligent methods for predicting electric smelting process parameters to improve the efficiency of mineral raw material processing

Introduction. Improving the efficiency of electric smelting processes remains an important challenge in mineral raw material processing because technological parameters are influenced by numerous nonlinear and interrelated factors. Conventional mathematical and statistical models often fail to provide sufficiently accurate predictions under continuously changing operating conditions. This study investigates the application of intelligent data-driven methods for forecasting electric smelting process parameters using historical industrial monitoring data and evaluates their suitability for real-time technological support. Methods. The proposed approach is based on archived process data collected from an automated process control system of an electric smelting furnace. Data preprocessing included noise filtering, missing-value interpolation, feature normalization, correlation analysis, and dimensionality reduction using principal component analysis. A Random Forest ensemble learning algorithm was developed and optimized by cross-validation. The predictive performance was evaluated using the coefficient of determination, mean absolute error, root mean square error, residual analysis, and robustness testing under different technological operating conditions. Results. The developed forecasting model demonstrated high predictive performance with a coefficient of determination of 0.962 on the independent test dataset and a mean absolute prediction error of 1.94%. More than 93% of all predictions were obtained with an error below 3%, while the average prediction time remained below 0.2 s. Feature importance analysis revealed that furnace active power, phase current, electrode position, and off-gas temperature provided more than 71% of the total predictive contribution. The model also maintained stable performance under measurement noise and during transient operating conditions, confirming its applicability to real industrial datasets. Conclusions. The obtained results demonstrate that the proposed intelligent forecasting approach provides accurate and computationally efficient prediction of electric smelting process parameters using only routinely collected industrial monitoring data. The developed model can be integrated into existing automated process control systems to improve process stability, support operational decision making, and increase the overall efficiency of mineral raw material processing. The proposed methodology also provides a basis for further development of intelligent digital technologies for metallurgical process control.

Correlation coefficient · Mean squared error · Principal component analysis · Process control · Random forest · Residual · Smelting · Iron and Steelmaking Processes · Metallurgical Processes and Thermodynamics · Molten salt chemistry and electrochemical processes