基于FCM-WPPCR协同度量的转炉炼钢终点碳温软测量方法

Soft sensor method for endpoint carbon temperature of BOF steelmaking based on FCM-WPPCR collaborative measure

中文关键词  转炉炼钢  预测分析  相似性度量  即时学习  软测量

英文关键词  BOF steelmaking  predictive analytics  similarity measurement  JITL  soft senser

中文摘要

      转炉炼钢终点碳温的准确检测是吹炼末期判断出钢的关键, 而非线性、多工况的数据特性是软测量有效预 测终点碳温的难点. 针对复杂工业数据下即时学习直接度量相关样本困难且度量依据较为单一, 致使样本匹配失 准的问题, 本文提出一种模糊C均值聚类加权后验概率与聚类关系(FCM-WPPCR)协同度量策略, 用于即时学习最 佳样本子集选择. 该策略通过模糊C均值聚类划分多工况子簇, 并引入推土机距离(EMD)准则构建一种后验概率模 式表征待测样本在各簇中的隶属度; 其次, 通过最大均值差异确定其他子簇与最大隶属度子簇的相关度后, 计算待 测样本与各簇样本EMD距离的均值构造一种协同度量机制加权后验概率和簇关系, 以确定待测样本在各簇中样本 选择的执行标准; 最后, 在各簇中选择相关样本构造最佳样本子集, 并建立局部回归模型预测终点碳温. 通过实际 炼钢生产过程数据仿真, 碳含量在±0.02%误差范围内预测精度达到92.60%, 温度在±10。C误差范围内预测精度达 到93.30%.

英文摘要

      The accurate detection of the endpoint carbon temperature in basic oxygen furnace (BOF) steelmaking is the key to determining the final stage of steelmaking, and the nonlinear and multiworking condition data characteristics are the difficulties in effectively predicting the endpoint carbon temperature using soft sensing. A fuzzy C-means clustering weighted posterior probability and cluster relation (FCM-WPPCR) co metric strategy is proposed to address the problem of inaccurate sample matching caused by the difficulty of directly measuring relevant samples in just-in-time learning (JITL) under complex industrial data and the relatively single measurement basis. This strategy is used to select the optimal sample subset for JITL. This strategy uses fuzzy C-means clustering to partition multicondition sub clusters, and introduces the Earth mover’s distance (EMD) criterion to construct a posterior probability pattern to characterize the membership degree of the tested sample in each cluster; Secondly, after determining the correlation between other sub clusters and the maximum membership sub cluster through the maximum mean difference, a collaborative measurement mechanism is constructed by calculating the mean of the EMD distance between the test sample and each cluster sample to construct a weighted posterior probability and cluster relationship, in order to determine the execution criteria for selecting the test sample in each cluster; Finally, selecting relevant samples from each cluster to construct the optimal sample subset, and establishing a local regression model to predict the end-point carbon temperature. Through simulation of actual steelmaking production process data, the prediction accuracy of carbon content reaches 92.60% within the error range of ±0.02%, and the prediction accuracy of temperature reaches 93.30% within the error range of ±10。C.