Matrix effects limit the quantitative performance of laser-induced breakdown spectroscopy (LIBS) for coal analysis. In this work, we proposed a data preprocessing method to reduce matrix effects, namely adaptive subset matching (ASM). ASM constructs a series of calibration models based on the similarity of sample matrix properties. Then an unknown sample is assigned to an appropriate model by calculating its matrix property. The proposed method was evaluated on 90 coal samples to determine the carbon content. Results demonstrate that ASM improves the quantitative performance of both multiple linear regression (MLR) and partial least square regression (PLSR). The root mean square error of prediction (RMSEP) is decreased from 6.19% to 3.23% for MLR and from 2.83% to 1.59% for PLSR, respectively. The corresponding mean pulse-to-pulse relative standard deviation (RSD) of prediction is decreased from 13.8% to 8.43% for MLR and from 4.59% to 2.48% for PLSR, respectively. Moreover, the results of calorific value, nitrogen and hydrogen quantification are improved. These results demonstrate that ASM can effectively reduce matrix effects for coal analysis using LIBS.

基质效应限制了用于煤分析的激光诱导击穿光谱法（LIBS）的定量性能。在这项工作中，我们提出了一种减少矩阵效应的数据预处理方法，即自适应子集匹配（ASM）。ASM根据样品基质属性的相似性构建一系列校准模型。然后，通过计算未知样本的矩阵属性，将其分配给适当的模型。在90个煤样上对提出的方法进行了评估，以确定碳含量。结果表明，ASM提高了多元线性回归（MLR）和偏最小二乘回归（PLSR）的定量性能。MLR的预测均方根误差（RMSEP）从6.19％降低至3.23％，PLSR的预测均方根误差从2.83％降低至1.59％。MLR的相应平均预测脉冲相对标准偏差（RSD）从13.8％降低至8.43％，PLSR的预测平均脉冲相对相对标准偏差（RSD）从4.59％降低至2.48％。此外，改进了热值，氮和氢定量的结果。这些结果表明，ASM可以有效减少使用LIBS进行煤炭分析的基质效应。