Optimization of Blast Furnace Ironmaking Using Machine Learning and Genetic Algorithms

Abstract

Blast furnace is a multiphase counter‐current packed bed reactor that converts iron‐bearing materials such as lumps, sinter, and pellets into hot metal using metallurgical coke and pulverized coal. The quality of input materials has a significant impact on furnace performance, hot metal quality and steel plant economics. It is difficult for operators to identify the optimal settings required for efficient and safe operation based on their experience alone, given the large number of furnace parameters. A multiobjective optimization problem for maximizing furnace productivity (PROD) and minimizing fuel rate (FR) with constraints on hot metal silicon (HMSi) and temperature (HMT) is formulated and solved using a genetic algorithm. Machine learning (ML) models are developed for PROD, FR, HMSi, and HMT and tested with data from an industrial blast furnace. Pareto‐optimal solutions along with optimal settings for key manipulated variables are obtained. It is demonstrated that PROD and FR can be improved by ≈3–5% at steady state. The overall ML model‐based optimization framework can be used as part of a blast furnace digital twin system to operate the furnace efficiently in real‐time for the given quality of raw materials.