将激光诱导击穿光谱技术（LIBS）应用于煤中热值的检测。 针对传统的通道面积归一化方法未能考虑煤质检测的物理/化学机制、 从而限制了所建模型在精确性、 准确性、 可重复性的情况, 提出了一种新型的基于光谱偏差产生原理的校正模型。 模型选取了19组煤样品, 随机选择其中15组为校正集, 用于建立热值的定量分析模型, 剩余四种为预测集, 用于对所建模型进行检验与评价。 模型从光谱偏差因素的产生因素出发, 通过原子光谱发射理论结合斯塔克展宽公式, 推导出LIBS条件下自吸收效应的影响机制及其所引起的偏差的修正方法。 通过元素间相互干扰结合基体效应的微观机理对基体效应进行光谱的偏差分析, 并根据K系数法的思路对LIBS中元素间相互干扰进行修正, 通过建立光谱的电子密度, 等离子体温度, 元素浓度的数值模型对基体效应引起的光谱偏差进行深度修正。 因而经过自吸收效应—元素间相互干扰—基体效应深度修正后, 模型对于所研究样品范围内其拟合优度R2=0.967, RMSEP=0.49 MJ·kg-1, RMSE=0.45 MJ·kg-1, MRE=2.42%, ARE=1.64%的同时RSD=5.79%, RSDP=8.10%。 相对于传统的通道面积归一化-多元线性回归方法的0.405, 8.28 MJ·kg-1, 4.14 MJ·kg-1, 22.85%, 52.48%, 18.28%, 32.85%, 表明测量的精确度与准确度都得到明显的提高, 证明该模型具有很好的应用价值。

A set of coal samples were used for laser-induced breakdown spectroscopy (LIBS) experiment to measure the coal calorific value. Traditional channel normalization method didn't consider the physical / chemical mechanism of coal, which would limit the model in precision, accuracy and repeatability. Thus a new calibrated model based on the kinds of the effects of spectral deviation was proposed in this paper. The model selected 19 groups of coal samples, where the random 15 groups were used to establish quantitative analysis model of calorific value while the remaining four for inspection and evaluation. The model based on spectral deviation factors, and the transmission theory combined with the stark broadening formula was used to deduce the absorption effect mechanism and the deviation correction method under the condition of LIBS. The mutual interference between elements and the mechanism of matrix effect were being analyzed while K coefficient method was used to correct mutual interference between the elements in the LIBS. The establishment of numerical model with the electron density, the plasma temperature and the element concentration was used to deeply corrected spectrum deviation caused by matrix effect. Thus taking into consideration of the effect of self-absorption, interfere of inter-elements and matrix effect, the calibration model was established, while R2=0.967, RMSEP=0.49 MJ·kg-1, RMSE=0.45 MJ·kg-1, MRE=2.42%, ARE=1.64%, RSD=5.79% and RSDP=8.10%. Compared with the 0.405, 8.28 MJ·kg-1, 4.14 MJ·kg-1, 22.85%, 52.48%, 18.28% and 32.85% of traditional channel normalized-multiple linear regression method, it demonstrated that the precision and accuracy have been improved significantly and model has good application value.