工程科学学报，第39卷，第4期：511-519,2017年4月 Chinese Journal of Engineering,Vol.39,No.4:511-519,April 2017 DOI:10.13374/j.issn2095-9389.2017.04.005:http://journals.ustb.edu.cn BP神经网络IF钢铝耗的预测模型 张思源”，包燕平)区，张超杰”，林路》 1)北京科技大学钢铁治金新技术国家重点实验室，北京1000832)钢铁研究总院治金工艺研究所，北京100081 区通信作者，E-mail:baoyp@usth.cdu.cn 摘要为了解决某钢厂F钢冶炼RH精炼过程铝耗偏高问题，通过数理统计和BP神经网络相结合的方法建立了铝耗预 测模型，并与多元线性回归模型进行比较，该模型具有更高准确度.该模型分析了不同治炼工艺参数对铝耗的具体影响，并 对相应工艺参数进行了优化.结果表明：脱碳结束氧活度或RH进站氧活度降低0.005%左右，每吨钢铝耗可降低0.07~ 0.08kg,铝脱氧有效利用系数为70.31%-80.35%：RH进站钢液温度增加35-40℃，铝耗降低1kg左右，铝热反应升温利用 系数在97.4%左右：吹氧量小于100m3和大于100m3时，氧气与铝反应的比例分别为37.3%和74.6%左右，吹氧量每增加 50m3,铝耗分别增加0.1kg和0.2kg左右.工艺参数优化后平均铝耗由1.359kg降低到1.113kg,降幅达18.1%. 关键词F钢：低碳钢：铝耗：神经网络：预测模型 分类号TF769.4 Prediction model of aluminum consumption with BP neural networks in IF steel production ZHANG Si-yuan,BAO Yan-ping,ZHANG Chao-jie,LIN Lu 1)State Key Laboratory of Advanced Metallurgy,University of Science and Technology Beijing,Beijing 100083,China 2)Metallurgical Technology Institute,Central Iron and Steel Research Institute,Beijing 100081,China Corresponding author,E-mail:baoyp@ustb.edu.cn ABSTRACT To solve the high aluminum consumption problem in interstitial-free steel production in a steel plant,an aluminum consumption prediction model was established by mathematical statistics and BP neural networks.Compared with the multiple linear regression model,this model's result is more accurate.The influence of different smelting processes on aluminum consumption was analyzed,and the process parameters were optimized.The results show that the amount of aluminum consumption per ton of steel decreases 0.07 to 0.08 kg when the oxygen activity before RH or after decarbonization reduces by 0.005%.The effective utilization coefficient of aluminum-deoxidizing is from 70.31%to 80.35%;the aluminum consumption decreases about 0.1 kg when the tempera- ture of steel before RH increases by 35 to 40 C.The heating utilization coefficient of aluminum thermal reaction is about 97.4%. When the blowing oxygen quantity is less than 100 m and greater than 100 m,the ratio of oxygen reacting with aluminum is about 37.3%or about 74.6%respectively,and the aluminum consumption increases by 0.1 kg or0.2kg,respectively,with the blowing ox- ygen quantity increasing by 50 m.After the process parameter optimization,the aluminum consumption decreases from 1.359 to 1.113 kg,which results in a decrease of 18.1%. KEY WORDS IF steel:low carbon steel:aluminum consumption:neural networks:prediction models F钢由于良好的深冲性能，被广泛应用在汽车和中夹杂物的数量、大小、形态和分布-园.F钢属于铝 家电行业中，对洁净度要求十分严格，因此需要控制钢 脱氧镇静钢，生产过程中要加入大量的铝合金进行脱 收稿日期：2016-07-25 基金项目：国家自然科学基金资助项目(51404022)：钢铁治金新技术国家重点实验室自主课题(41616003)工程科学学报，第 39 卷，第 4 期: 511--519，2017 年 4 月 Chinese Journal of Engineering，Vol． 39，No． 4: 511--519，April 2017 DOI: 10． 13374 /j． issn2095--9389． 2017． 04． 005; http: / /journals． ustb． edu． cn BP 神经网络 IF 钢铝耗的预测模型 张思源1) ，包燕平1) ，张超杰1) ，林 路2) 1) 北京科技大学钢铁冶金新技术国家重点实验室，北京 100083 2) 钢铁研究总院冶金工艺研究所，北京 100081 通信作者，E-mail: baoyp@ ustb． edu． cn 摘 要 为了解决某钢厂 IF 钢冶炼 ＲH 精炼过程铝耗偏高问题，通过数理统计和 BP 神经网络相结合的方法建立了铝耗预 测模型，并与多元线性回归模型进行比较，该模型具有更高准确度． 该模型分析了不同冶炼工艺参数对铝耗的具体影响，并 对相应工艺参数进行了优化． 结果表明: 脱碳结束氧活度或 ＲH 进站氧活度降低 0. 005% 左右，每吨钢铝耗可降低 0. 07 ～ 0. 08 kg，铝脱氧有效利用系数为 70. 31% ～ 80. 35% ; ＲH 进站钢液温度增加 35 ～ 40 ℃，铝耗降低 1 kg 左右，铝热反应升温利用 系数在 97. 4% 左右; 吹氧量小于 100 m3 和大于 100 m3 时，氧气与铝反应的比例分别为 37. 3% 和 74. 6% 左右，吹氧量每增加 50 m3 ，铝耗分别增加 0. 1 kg 和 0. 2 kg 左右． 工艺参数优化后平均铝耗由 1. 359 kg 降低到 1. 113 kg，降幅达 18. 1% ． 关键词 IF 钢; 低碳钢; 铝耗; 神经网络; 预测模型 分类号 TF769. 4 收稿日期: 2016--07--25 基金项目: 国家自然科学基金资助项目( 51404022) ; 钢铁冶金新技术国家重点实验室自主课题( 41616003) Prediction model of aluminum consumption with BP neural networks in IF steel production ZHANG Si-yuan1) ，BAO Yan-ping1)  ，ZHANG Chao-jie1) ，LIN Lu2) 1) State Key Laboratory of Advanced Metallurgy，University of Science and Technology Beijing，Beijing 100083，China 2) Metallurgical Technology Institute，Central Iron and Steel Ｒesearch Institute，Beijing 100081，China Corresponding author，E-mail: baoyp@ ustb． edu． cn ABSTＲACT To solve the high aluminum consumption problem in interstitial-free steel production in a steel plant，an aluminum consumption prediction model was established by mathematical statistics and BP neural networks． Compared with the multiple linear regression model，this model's result is more accurate． The influence of different smelting processes on aluminum consumption was analyzed，and the process parameters were optimized． The results show that the amount of aluminum consumption per ton of steel decreases 0. 07 to 0. 08 kg when the oxygen activity before ＲH or after decarbonization reduces by 0. 005% ． The effective utilization coefficient of aluminum-deoxidizing is from 70. 31% to 80. 35% ; the aluminum consumption decreases about 0. 1 kg when the tempera￾ture of steel before ＲH increases by 35 to 40 ℃ ． The heating utilization coefficient of aluminum thermal reaction is about 97. 4% ． When the blowing oxygen quantity is less than 100 m3 and greater than 100 m3 ，the ratio of oxygen reacting with aluminum is about 37. 3% or about 74. 6% respectively，and the aluminum consumption increases by 0. 1 kg or 0. 2 kg，respectively，with the blowing ox￾ygen quantity increasing by 50 m3 ． After the process parameter optimization，the aluminum consumption decreases from 1. 359 to 1. 113 kg，which results in a decrease of 18. 1% ． KEY WOＲDS IF steel; low carbon steel; aluminum consumption; neural networks; prediction models IF 钢由于良好的深冲性能，被广泛应用在汽车和 家电行业中，对洁净度要求十分严格，因此需要控制钢 中夹杂物的数量、大小、形态和分布［1--2］． IF 钢属于铝 脱氧镇静钢，生产过程中要加入大量的铝合金进行脱