样本量对激光诱导击穿光谱技术在煤质灰分检测的影响研究  

Effects of sample size on coal ash analysis based on laser-induced breakdown spectroscopy

作　　者：倪浩然 周清 朱秋阁 李衍方 崔小峰 成中豪 倪琳 NI Hao-ran;ZHOU Qing;ZHU Qiu-ge;LI Yan-fang;CUI Xiao-feng;ZHENG Zhong-hao;NI Lin(Xuzhou Inspection and Testing Center,Xuzhou221000,China;SuZhou Golden Concord Energv Technology Co,Ltd,Suzhou 215000,China;Beijing Collaboration Xinguang Detection Technology Co.,Ltd,Beijing 100194,China)

机构地区：[1]徐州市检验检测中心,徐州221000 [2]苏州协鑫能源科技有限公司,苏州215000 [3]北京协同鑫光检测技术有限公司,北京100195

出　　处：《煤炭加工与综合利用》2022年第12期86-91,共6页Coal Processing & Comprehensive Utilization

基　　金：江苏省市场监督管理局科研项目KJ21125054《基于激光诱导击穿光谱技术煤质快检方法研究及仪器开发》。

摘　　要：为了提高激光诱导击穿光谱技术在定量检测煤质灰分时的精度,对同一煤矿的77个标准煤样进行光谱数据采集,使用偏最小二乘法(PLS)对不同数量的样本数据进行建模,再用剩余样品进行预测验证。实验结果表明:基于偏最小二乘法的灰分预测模型在训练样本数量小于60个时,建模精度较差,合格率约为35%,随着训练样本的增加,预测精度逐渐提高,当样本量达到67个时,预测精度基本稳定,合格率为90%,当训练样本数量为70个时,预测精度最佳,合格率为100%,仅有一个预测偏差大于0.6%,继续增加训练样本至76个时,预测精度变差,合格率变为83%,出现过拟合问题,当训练样本的数量不变时,增加单个样本的光谱数据,可显著提高模型的预测精度。研究结果表明,基于偏最小二乘回归法的定量分析模型,可根据预测结果选取合适的训练样本数量,构建出精度较高的煤质灰分预测模型,为进一步研发检测仪器提供技术支撑。To improve the accuracy of laser induced breakdown spectroscopy in the quantitative detection of coal quality,spectral data were collected from 77 standard coal samples from the same coal mine.Partial least squares(PLS) method is used to model different number of sample data,and then the remaining samples are used for prediction verification.The results show that the modeling accuracy of the ash prediction model based on partial least squares method is poor when the number of training samples is less than 60,and the qualified rate is about 35%.With the increase of training samples,the prediction accuracy gradually improves.Prediction accuracy when the sample size to 67,basically stable,while the number of training samples is 70,best precision,percent of pass is 100%,only one forecast deviation is greater than 0.6%,and continue to increase the training sample to 76 prediction accuracy,the percent of pass is 61%,there may be a fitting.When the number of training samples is unchanged,adding the spectral data of similar samples can significantly improve the prediction accuracy of the model.The research shows that the quantitative analysis model based on partial least squares regression method can select the appropriate number of training samples according to the prediction results,build a high accuracy coal ash prediction model and provide a technical possibility for further research on detection instruments.

关 键 词：激光诱导击穿光谱技术 煤灰分 偏最小二乘回归法 样本量 定量分析 

分 类 号：TN247[电子电信—物理电子学]