PCR和PLSR在煤质全水分检测中的比较研究  

Comparative study of PCR and PLSR modeling methods in coal moisture detection

作　　者：宁石茂 NING Shi-mao(Tunlan Coal Preparation Plant,Xishan Coal Electricity Group Co.,LTD,Gujiao,Shanxi 030206,China)

机构地区：[1]西山煤电(集团)有限责任公司屯兰矿选煤厂,山西古交030206

出　　处：《煤炭加工与综合利用》2023年第10期60-63,共4页Coal Processing & Comprehensive Utilization

摘　　要：为了快速、无损的检测出煤质内部的全水分含量,研究采集了200个焦煤样品的近红外光谱,采用马氏距离和学生式残差相结合的方法剔除了异常样品,并对其进行了一阶微分、二阶微分、15点平滑、多元散射校正(MSC)和标准归一化处理(SNV)光谱预处理,采用主成分回归(PCR)和偏最小二乘回归(PLSR)对煤样进行建模分析。试验结果表明:经SNV预处理后的PCR模型最佳,校正集和交叉验证集相关系数分别为0.903和0.874,均方根误差分别为0.089和0.132;经15点平滑处理后的PLSR模型最佳,校正集和交叉验证集相关系数分别为0.974和0.887,均方根误差分别为0.038和0.043。PLSR模型相比PCR模型更具有代表性,模型稳定性和预测能力更强。In order to quickly and nondestructive detect the total moisture content in the coal,the near-infrared spectra of 200 bituminous coal samples were collected in this paper,the method of combining Markov distance with student residual is used to eliminate abnormal samples,first order differential,second order differential,15 points smoothing,MSC and SNV spectral preprocessing were performed,PCR and PLSR were used to model and analyze the coal samples.The results indicated that PCR model after MSC pretreatment is the best,the correlation coefficients of correction set and cross validation set are 0.938 and 0.746,and the root mean square errors are 0.075 and 0.103 respectively.PLSR model after 15 point smoothing is the best,the correlation coefficients of correction set and cross validation set are 0.974 and 0.887,and the root mean square errors are 0.038 and 0.043 respectively.Compared with PCR model,PLSR model is more representative,stable and predictive.

关 键 词：近红外光谱 主成分回归 偏最小二乘回归 全水分 光谱预处理 

分 类 号：TQ533.6[化学工程—煤化学工程]