煤中全水分及灰分的近红外测试方法研究

Near infrared test method of total moisture and ash in coal

为了快速、无损检测煤质中的全水分和灰分,采集了120个精煤样品的近红外漫反射光谱,对微分光谱进行分析,利用偏最小二乘回归(PLSR)算法结合不同光谱预处理方法建立基于马氏距离剔除异常样品后的定量数学模型,分析预测值与真实值的相关性,并对最优预处理下的模型残差进行讨论。结果表明:经过多元散射校正处理后建立的全水分模型效果最优,相关系数达到0.982 12,校正集均方根误差为0.013,预测集均方根误差为0.017。经过5点平滑预处理后建立的灰分模型效果最佳,相关系数达到0.947 47,校正集均方根误差为0.058,预测集均方根误差为0.052,2项指标的残差波动均匀,模型的稳定性和预测能力较强。

In order to detect total moisture and ash of coal quickly and with lossless,120 near-infrared diffuse diffuse reflection spectrums were collected from 120 clean coal samples. Differential spectra were analyzed. Based on markov distance to eliminate abnormal samples,a mathematical model of quantitative was estabilished combining partial least squares regression（PLSR） and different spectral preprocessing methods. The correlation between predicted values and actual values was analyzed. The residual error under the best pretreatment model was analyzed. The results indicated that modeling of moisture through multiple scattering correction（MSC） had better effects,the correlation coefficient was 0. 982 12,the root mean square error of calibration（RMSEC） and root mean square error of prediction（RMSEP）were 0. 013 and 0. 017. The modeling of ash through 5 points smooth had a better result,the correlation coefficient was 0. 947 47,the root mean square error of calibration（RMSEC） and root mean square error of prediction（RMSEP） were 0. 058 and 0. 052,the residual error of two indicators waved in uniform,and the model had better stability and predictive ability.