利用近红外光谱分析技术的诸多优势, 可实现煤质多项指标的快速分析。 煤样光谱数据的采集易受到噪声散射的干扰, 在建模分析前常用多元散射校正方法对其予以恢复去噪。 而传统的多元散射校正方法存在以线性表达式描述非线性关系的局限性, 以及盲目追随“理想”光谱而导致的失真问题, 为此提出基于拟线性局部加权法的煤样光谱散射校正。 该法先选取二次曲线、 三次曲线和增长曲线等三种拟线性函数, 代替原线性函数; 再分别引入均值核函数、 高斯核函数、 Epanechnikov核函数、 二次权重核函数及三次权重核函数等五种核函数, 在原评估函数中构造局部加权函数, 利用局部加权函数精确表述各波长点处的依赖关系; 最后, 分析不同模式下校正光谱的准确性, 确定拟线性函数与核函数的最佳组合。 为验证方法的适用性, 对0.2, 1和3 mm等三种粒度等级下的煤样光谱数据进行散射校正。 结果表明: 改进多元散射校正方法在消除光谱中散射干扰的同时, 放大了特征谱峰信息, 校正光谱与煤样各项指标的相关性得到了显著增强, 有效地提高了煤质近红外光谱分析模型的预测精度和稳定性。

The present paper puts forth a new spectral correction method based on quasi-linear expression and local weighted function. The first stage of the method is to search 3 quasi-linear expressions to replace the original linear expression in MSC method, such as quadratic, cubic and growth curve expression. Then the local weighted function is constructed by introducing 4 kernel functions, such as Gaussian, Epanechnikov, Biweight and Triweight kernel function. After adding the function in the basic estimation equation, the dependency between the original and ideal spectra is described more accurately and meticulously at each wavelength point. Furthermore, two analytical models were established respectively based on PLS and PCA-BP neural network method, which can be used for estimating the accuracy of corrected spectra. At last, the optimal correction mode was determined by the analytical results with different combination of quasi-linear expression and local weighted function. The spectra of the same coal sample have different noise ratios while the coal sample was prepared under different particle sizes. To validate the effectiveness of this method, the experiment analyzed the correction results of 3 spectral data sets with the particle sizes of 0.2, 1 and 3 mm. The results show that the proposed method can eliminate the scattering influence, and also can enhance the information of spectral peaks. This paper proves a more efficient way to enhance the correlation between corrected spectra and coal qualities significantly, and improve the accuracy and stability of the analytical model substantially.