Enhanced steelmaking cost optimization and real-time alloying element yield prediction: a ferroalloy model based on machine learning and linear programming

Abstract

The production of ferroalloys is a resource-intensive and energy-consuming process. To mitigate its adverse environmental effects, steel companies should implement a range of measures aiming at enhancing the utilization rate of ferroalloys. Therefore, a comprehensive ferroalloy model was proposed, incorporating a prediction model for alloying element yield based on case-based reasoning and support vector machine (CBR–SVM), along with a ferroalloy batching model employing an integral linear programming algorithm. In simulation calculations, the prediction model exhibited exceptional predictive performance, with a hit rate of 96.05% within 5%. The linear programming ingredient model proved effective in reducing costs by 20.7%, which was achieved through accurate adjustments to the types and quantities of ferroalloys. The proposed method and system were successfully implemented in the actual production environment of a specific steel plant, operating seamlessly for six months. This implementation has notably increased the product quality of the enterprise, with the control rate of high-quality products increasing from 46% to 79%, effectively diminishing the consumption and expenses associated with ferroalloys. The reduced usage of ferroalloys simultaneously reduces energy consumption and mitigates the adverse environmental impact of the steel industry.