摘要
高炉热状态是影响高炉冶炼的重要因素之一,对其进行监控和预测是业界关注的重点。高炉热状态的预测常转化为对铁水硅含量数值变化趋势进行预测的分类问题,以及对铁水硅含量具体数值进行预测的回归问题。结合长短期记忆(LSTM)和多任务学习提出了一种神经网络模型,用于同时解决高炉热状态预测的分类和回归问题。模型在两个典型的高炉数据集上进行训练,分类预测结果准确率可达0.84,回归预测结果命中率和相关系数分别可达0.76和0.819 6。训练了两种单任务学习模型并与多任务学习模型进行对比。试验结果表明,多任务学习模型能同时提升高炉热状态分类和回归任务的预测性能,预测结果均优于单任务模型。
The thermal state of blast furnace is one of the important factors affecting the blast furnace smelting,and its monitoring and prediction are the focus of the industry.The prediction of blast furnace thermal state is often transformed into the classification problem which focuses on predicting the change trend of silicon content in hot metal,and the regression problem which focuses on predicting the specific value of silicon content in hot metal.A neural network model was proposed combined with long short-term memory(LSTM)and multi-task learning to solve the classification and regression problems of blast furnace thermal state prediction at the same time.The model was trained on two typical blast furnace data sets,and the classification prediction result could reach the accuracy of 0.84,and the regression prediction result could reach the hit rate of 0.76 and the correlation coefficient of 0.8196.Two single task learning models were trained and compared with multi-task learning models.The experimental results show that the multi-task learning model can simultaneously improve the prediction performance of blast furnace thermal state classification and regression tasks,and the prediction results are better than the single task model.
作者
胡进
郜传厚
HU Jin;GAO Chuanhou(Polytechnic Institute,Zhejiang University,Hangzhou 310015,Zhejiang,China;School of Mathematical Science,Zhejiang University,Hangzhou 310058,Zhejiang,China)
出处
《中国冶金》
CAS
CSCD
北大核心
2023年第7期81-90,共10页
China Metallurgy
基金
国家自然科学基金资助项目(12071428,62111530247)
浙江省自然科学基金重点项目(LZ20A010002)。
关键词
高炉
热状态预测
铁水硅含量
长短期记忆网络
多任务学习
blast furnace
thermal state prediction
silicon content in hot metal
LSTM
multi-task learning
