|国家科技期刊平台
首页|期刊导航|中南大学学报(自然科学版)|基于卷积神经网络的烧结成品率预测

基于卷积神经网络的烧结成品率预测OA北大核心CSTPCD

Sintering yield prediction based on convolutional neural network

中文摘要英文摘要

成品率是综合反映烧结矿产量、质量和能耗的关键性指标.针对成品率检测存在滞后性问题,以烧结机尾断面图像作为输入,通过卷积神经网络拟合断面图像与成品率的关系,实现烧结成品率的在线预测.根据烧结机尾断面红外图像的特点,采用DenseNet网络结构的卷积神经网络作为建模方法,并以多尺度稠密连接块对网络结构进行改进,在同一层网络中提取多尺度、图像信息,通过拟合高维图像特征与烧结成品率之间的关系,实现成品率的准确预测.采用国内某大型钢铁生产企业的机尾断面图像和成品率数据对模型进行验证.研究结果表明:所提出的改进DenseNet网络在烧结成品率预测问题上具有较强的拟合和泛化能力,定义成品率预测值与实际值绝对误差在±2.8%的区间范围内为命中目标,模型命中率达92.66%,且均方根误差仅为1.76%,可为生产工艺参数的优化调控提供依据.

Sintering yield is a comprehensive indicator that reflects the production,quality and energy consumption in sintering processes.Addressing the lag in sintering yield detection,the cross-sectional image of the sintering machine tail was used as input and a convolutional neural network(CNN)was employed to fit the relationship between the cross-sectional image and sintering yield,thereby online prediction of sintering yield could be realized.According to the characteristics of infrared images of sintering tail section,a CNN with a DenseNet architecture was utilized for modeling.The network structure was enhanced with multi-scale dense blocks to extract information from images of various sizes within the same layer.By fitting the relationship between high-dimensional image feature information and sintering yield,accurate predictions of the sintering yield were achieved.The model was validated by using historical cross-sectional images and sintering yield data from a major domestic steel production company.The results show that the proposed improved DenseNet network had strong fitting and generalization ability in the sintering yield prediction problem.Defining the absolute error between the predicted and actual yield values in the interval range of±2.8%as the hitting target,the model hitting rate reached 92.66%,and the root mean sguare error was only 1.76%,which could provide a basis for the optimization of the production process parameters.

彭梓塘;黄晓贤;范晓慧;赵利明;李骞;陈许玲;匡朝辉;甘敏

中南大学资源加工与生物工程学院,湖南长沙,410083宝钢湛江钢铁有限公司炼铁厂,广东湛江,524000

冶金工业

烧结过程成品率预测断面图像卷积神经网络

sintering processsintering yield predictioncross-sectional imageconvolutional neural network

《中南大学学报(自然科学版)》 2024 (004)

1263-1271 / 9

国家自然科学基金资助项目(52274344)(Project(52274344)supported by the National Natural Science Foundation of China)

10.11817/j.issn.1672-7207.2024.04.001

评论