基于长短期时间序列网络的氧气需求量预测  

Oxygen consumption prediction with long short-term time-series network

作　　者：常婷婷[1] 张颜颜 澈格乐根 刘洪屾 CHANG Tingting;ZHANG Yanyan;CHE Gelegen;LIU Hongshen(Energy and Environment Division,Shanghai Baosight Software Co.,Ltd.,Shanghai 201203,China;Key Laboratory of Data Analytics and Optimization for Smart Industry(Northeastern University),Ministry of Education,Shenyang 110819,China;Liaoning Engineering Laboratory of Data Analytics and Optimization for Smart Industry,Shenyang 110819,China;Liaoning Key Laboratory of Manufacturing System and Logistics Optimization,Shenyang 110819,China)

机构地区：[1]上海宝信软件股份有限公司能环事业部,上海201203 [2]智能工业数据解析与优化教育部重点实验室(东北大学),辽宁沈阳110819 [3]辽宁省智能工业数据解析与优化工程实验室,辽宁沈阳110819 [4]辽宁省制造系统与物流优化重点实验室,辽宁沈阳110819

出　　处：《冶金自动化》2022年第3期25-33,共9页Metallurgical Industry Automation

基　　金：国家自然科学基金重点国际合作项目(71520107004);国家自然科学基金重大项目(71790614);高等学校学科创新引智计划(111计划)(B16009)。

摘　　要：钢铁企业转炉炼钢过程中的氧气消耗机理复杂,具有非线性、间歇性等特点。采用深度学习方法分析炼钢用氧规律,对未来一个炼钢吹炼计划内的氧气需求量进行多步预测,获取分钟级耗氧量及趋势曲线。针对工业数据中的异常值与缺失值进行预处理,建立基于超参数调节的长短期网络(long short-term network,LSTNet)预测模型,采用钢铁企业炼钢过程用氧量的实际数据对算法进行测试,并与基于长短期记忆网络(long short-term memory,LSTM)的预测模型对比,相比LSTM预测模型,平均绝对误差(mean absolute error,MAE)指标降低了12.3%,均方根误差(root mean square error,RMSE)指标则降低了8.5%,表明该预测方法对钢铁企业转炉炼钢过程中氧气需求量的预测效果较好。The mechanism of oxygen consumption in the process of converter steelmaking in iron and steel enterprises is complex,and which has the characteristics of nonlinearity and intermittence.The law of oxygen consumption in steelmaking was analyzed by deep learning method to carry out multi-step prediction of the minute-level oxygen demand of a steelmaking plan in the future.Firstly,the outliers and missing values in the industrial data were preprocessed,and the long short-term network(LSTNet)model based on super parameter adjustment was established to predict the oxygen demand in the future planning period.The actual data of oxygen consumption in steelmaking process was used to test the proposed algorithm,and compared with the long short-term memory(LSTM)model.The results show that the MAE is reduced by 12.3%,and the RMSE is reduced by 8.5%,indicating that LSTNet model has a better effect on oxygen demand prediction in converter steelmaking process of iron and steel enterprises.

关 键 词：氧气系统 深度学习 预测模型 转炉炼钢 长短期网络 长短期记忆网络 

分 类 号：TF713[冶金工程—钢铁冶金] TP18[自动化与计算机技术—控制理论与控制工程]