当前位置：首页 > 过刊浏览->2025年46卷第10期
基于贝叶斯优化机器学习的转炉耗氧量预测研究
Researchon the Predictionof Oxygen Consumptionin Convertersvia Bayesian-optimizedMachineLearning
浏览(21) 下载(1)
- DOI:
- 作者:
- 丁志豪 1，信自成 1, 2，张江山 1，刘青 1
DING Zhihao1,XIN Zicheng1, 2,ZHANG Jiangshan1,LIU Qing1
- 作者单位:
- 1. 北京科技大学 绿色低碳钢铁冶金全国重点实验室，北京 100083；2. 北京科技大学 自动化学院，北京 100083
1. State Key Laboratory of Advanced Metallurgy, University of Science and Technology Beijing, Beijing 100083, China; 2. School of Automation and Electrical Engineering, University of Science and Technology Beijing, Beijing 100083,China
- 关键词:
- 转炉；耗氧量预测；BP 神经网络算法；极限学习机；贝叶斯优化
converter; oxygen consumption prediction; BP neural network algorithm; extreme learning machine; Bayesian optimization
- 摘要:
- 转炉冶炼过程具有多变量、非线性、强耦合的特点，冶炼过程吹氧控制对钢液成分和温度具有重要影响。 为实现转炉吹氧量的精确预测， 利用箱线图法对实际生产数据进行预处理， 基于反向传播神经网络 (back propagation neuralntework, BP)算法和极限学习机(extreme learning machine, ELM)算法，构建了转炉耗氧量预测模型，运用贝叶斯优化(bayesian optimization, BO)算法，对 BP 神经网络算法和 ELM 算法的超参数进行寻优；最后，采用多种评价指标对所建模型性能进行了评估 。 结果 表 明 ，BO-ELM 预测 模 型 性 能 优 于 BO-BP 预测 模 型 ，BO-ELM 耗氧 量 预 测 模 型 的 R2、RMSE 和 MAE 分别为 0.721、137.176 和 113.622，且耗氧量在±300 m3 误差范围内的命中率达 98.10%。The converter smelting process is characterized by multivariable, nonlinear, and strongly coupled dynamics, where oxygen blowing control significantly influences the composition and temperature of molten steel. To achieve precise forecasting of the oxygen-blowing volume, actual production data were first preprocessed via the boxplot method. Subsequently, prediction models for converter oxygen consumption were constructed on the basis of the back propagation neural network (BP) algorithm and the extreme learning machine (ELM) algorithm. The Bayesian optimization (BO) algorithm was employed to optimize the hyperparameters of the BP neural network algorithm and ELM algorithm. Finally, model performance was evaluated via multiple metrics. The results demonstrate that the BO-ELM prediction model outperforms the BO-BP model, achieving R2,RMSE, and MAE values of 0.721, 137.176,and 113.622,respectively. The hit ratio within the error range of ±300 m3 of oxygen consumption was 98.10%.