基于CNN-LSTM的水泥熟料f-CaO预测模型  

Cement Clinker Free Calcium Oxide Prediction Model Based on CNN-LSTM

作　　者：郑涛[1,2] 刘辉 陈薇[1,2] 杨恺[1,2] 张建飞 褚彪 ZHENG Tao;LIU Hui;CHEN Wei;YANG Kai;ZHANG Jianfei;CHU Biao(School of Electrical Engineering and Automation,Hefei University of Technology,Hefei 230009,China;Anhui Engineering Technology Research Center of Industrial Automation,Hefei University of Technology,Hefei 230009,China;Hefei Cement Research and Design Institute Co.,Ltd.,Hefei 230051,China)

机构地区：[1]合肥工业大学电气与自动化工程学院,安徽合肥230009 [2]合肥工业大学安徽省工业自动化工程技术研究中心,安徽合肥230009 [3]合肥水泥研究设计院有限公司,安徽合肥230051

出　　处：《控制工程》2024年第7期1263-1271,共9页Control Engineering of China

基　　金：安徽省重点研发计划项目(202104a05020054);青年教师科研创新启动专项A项目(JZ2021HGQ0195)。

摘　　要：水泥熟料中游离氧化钙(f-CaO)含量的传统人工离线检测缺乏时效性,不利于生产指导。针对离线检测的滞后问题和软测量模型中f-CaO含量与辅助变量的时序匹配问题,提出了一种基于卷积神经网络(convolutional neural network,CNN)和长短时记忆(long short-term memory,LSTM)神经网络的f-CaO含量预测模型。首先,利用滑动窗口截取辅助变量的区间数据;然后,采用CNN提取区间数据的时序特征;之后,构建LSTM神经网络模型;最后,控制截取辅助变量的延迟时间和间隔时间,根据模型预测拟合度提取辅助变量的最优时序特征。仿真结果表明,所提模型提高了水泥熟料中f-CaO含量的预测精度。The traditional manual offline detection of free calcium oxide(f-CaO)content in cement clinker lacks timeliness,which is not conducive to production guidance.To deal with the hysteresis problem in offline detection and the timing matching problem between f-CaO content and auxiliary variables in the soft measurement model,the f-CaO content prediction model based on the convolutional neural network(CNN)and long short-term memory(LSTM)neural network is proposed.Firstly,the interval data of auxiliary variables are intercepted by sliding window.Secondly,the CNN is used to extract the temporal features of the interval data.Thirdly,the LSTM neural network is constructed.Finally,the delay time and interval time of the interception in auxiliary variables are controlled,and the optimal temporal features of auxiliary variables are extracted according to the model prediction fit.The simulation results show that the proposed model can improve the prediction accuracy of f-CaO content in cement clinker.

关 键 词：时序特征 滑动窗口 CNN LSTM神经网络 最优时序特征 预测精度 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]