Scholarlink

<p>A novel hybrid machine learning and optimization framework is proposed to reduce the specific energy consumption (SEC) in Electric Arc Furnace (EAF) steelmaking via the direct reduced iron (DRI) route. Three predictive models Artificial Neural Networks (ANN), Genetic Programming (GP), and Multi-Gene Genetic Programming (MGGP) were developed using 22 input features representing key operational and material variables. Model validation employed over 8000 industrial heats alongside a physics-informed truth table framework grounded in known EAF behavior. The ANN model achieved the highest prediction accuracy (<i>R</i><sup>2</sup> = 0.95, RMSE = 7.78), while MGGP balanced accuracy (<i>R</i><sup>2</sup> = 0.86) with interpretability through symbolic equations. The GP model yielded lower performance (<i>R</i><sup>2</sup> = 0.65). These models were coupled with a Genetic Algorithm (GA) to optimize two operator-controlled inputs oxygen injection and tapping temperature while carbon follows a fixed ratio to oxygen on a batch-wise basis. GA optimization reduced average SEC from 412 to 390&#xa0;kWh/ton, enabling an energy saving of 22&#xa0;kWh/ton. For a steel plant producing 1.7 million tons annually, this implies a potential CO₂ reduction of approximately 0.017 million tons per year. The proposed hybrid, explainable, and physically validated approach offers a robust and scalable pathway toward operational efficiency and decarbonization in EAF-based steelmaking.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid Explainable Machine Learning and Optimization Framework for Energy Reduction in DRI-Based Electric Arc Furnace Steelmaking

  • 期刊论文

  • 发表日期：

    2025年10月6日

查看全文
  • Narottam Behera,
  • Hany Hamed,
  • Arnab Ghosh,
  • Chhavi Sharma,
  • Sandip Kumar Lahiri

摘要

A novel hybrid machine learning and optimization framework is proposed to reduce the specific energy consumption (SEC) in Electric Arc Furnace (EAF) steelmaking via the direct reduced iron (DRI) route. Three predictive models Artificial Neural Networks (ANN), Genetic Programming (GP), and Multi-Gene Genetic Programming (MGGP) were developed using 22 input features representing key operational and material variables. Model validation employed over 8000 industrial heats alongside a physics-informed truth table framework grounded in known EAF behavior. The ANN model achieved the highest prediction accuracy (R2 = 0.95, RMSE = 7.78), while MGGP balanced accuracy (R2 = 0.86) with interpretability through symbolic equations. The GP model yielded lower performance (R2 = 0.65). These models were coupled with a Genetic Algorithm (GA) to optimize two operator-controlled inputs oxygen injection and tapping temperature while carbon follows a fixed ratio to oxygen on a batch-wise basis. GA optimization reduced average SEC from 412 to 390 kWh/ton, enabling an energy saving of 22 kWh/ton. For a steel plant producing 1.7 million tons annually, this implies a potential CO₂ reduction of approximately 0.017 million tons per year. The proposed hybrid, explainable, and physically validated approach offers a robust and scalable pathway toward operational efficiency and decarbonization in EAF-based steelmaking.

Graphical Abstract