热值是煤质特性的重要参数之一, 很大程度上影响着燃煤锅炉的运行。 为了克服传统检测方法所存在的问题, 将激光诱导击穿光谱(LIBS)应用于燃煤热值的定量分析。 煤的结构复杂, 所含的元素种类众多, 包括了主量元素、 次量元素和痕量元素, 致使煤的LIBS光谱信息复杂。 如何有效提取LIBS光谱信息, 实现准确的定量化测量是LIBS在煤特性检测中发挥作用的前提和基础。 近年来, 随着人工智能技术的发展, 相关的分析技术也开始应用于煤的工业指标分析和热值预测中。 为实现煤样品中LIBS光谱信息的有效提取, 同时为克服常规的分析方法易出现的过渡拟合、 收敛性不好等问题, 提出采用结合K-fold Cross Validation(K-CV)参数优化的支持向量机(SVM)回归方法, 实现LIBS定量分析煤中的热值。 SVM方法是结构风险最小化的近似实现, 可用于模式分类和非线性回归。 为了得到有效的LIBS分析模型, 实验选用44种电厂常用的热值含量不同的煤样作为实验对象, 选择其中33个作为训练集, 剩余11个为测试集。 利用搭建的LIBS实验系统获取所选煤样品的等离子体发射光谱数据, 首先分析了SVM热值回归模型的参数-惩罚因子C、 核函数参数g与模型精度的关联, 确定C和g最佳取值范围, 然后分别建立了基于LIBS全谱和某些元素(非金属元素和金属元素)特征光谱的SVM回归模型。 利用训练集光谱数据, 结合K-CV法得到热值SVM回归模型的最优参数C和g的值, 建立基于SVM最优参数的煤热值定量分析模型。 然后将测试集的光谱数据作为输入量用于测试所建立模型的可靠性, 得到分别采用全谱、 非金属元素特征光谱、 非金属与金属元素特征谱相结合的热值定量分析模型, 其决定系数R2均达到0.99以上, 均方误差分别为0.12, 0.17和0.06 (MJ·kg-1)2, 预测平均相对偏差分别为1.2%, 1.23%和0.69%。 结果表明： 基于K-CV参数优化SVM回归方法可用于LIBS技术实现燃煤热值的定量分析, 且可得到较高的分析精确度和准确度; 同时通过对比选用不同的光谱特征的定量分析模型可知, 采用非金属与金属元素的特征光谱所建立的基于K-CV参数优化SVM的热值定量模型, 能够有效提高LIBS应用于快速检测煤热值的精度和准确度, 实现煤热值的准确预测。

Heat value is one of the important parameters of coal quality, and it greatly affects the operation of coal-fired boilers. In order to overcome the shortcomings of traditional detection methods, laser induced breakdown spectroscopy (LIBS) was applied to quantitative analysis of heat value in coal. The structure of coal is complex. It contains many types of elements, including major, minor and trace elements, which would result the complexity of LIBS spectral information from coal. It is premise and foundation to effectively extract LIBS spectral information and achieve accurate quantitative measurement by using LIBS. In recent years, as the development of artificial intelligence technology, relevant analytical techniques have also been applied to the proximate analysis and heat value prediction of coal. In order to realize the effective extraction of LIBS spectral information in coal samples and overcome the problems such as transient fitting and poor convergence that are easily caused by conventional analytical methods, the K-fold cross validation (K-CV) parameter-optimized combined with Support Vector Machine (SVM) regression was proposed to quantitatively analyze the heat value in coal. The SVM method is an approximate realization of structural risk minimization, which can be used for pattern classification and nonlinear regression. 44 coal samples with different heat values commonly used in the power plant were selected as experimental objects, 33 of which were selected as training sets, and the remaining 11 were test sets. The correlation between the parameters of the SVM regression model-penalty factor C, the kernel function parameter g and the model accuracy were firstly analyzed based on the laser-induced coal spectrum and the best search scope for C and g were determined. Then the SVM regression model was established based on the LIBS full-spectrum and some typical elements (non-metallic elements and metal elements) feature spectra, respectively. The optimal parameters C, g of the heat value SVM regression model is obtained by using the training set spectral data, combined with the K-CV method. The spectral features of the prediction set as input are to test the reliability of the model. The calibration model established by the full spectrum, non-metallic element characteristic spectrum, as well as non-metal and metal element characteristic spectrum, respectively, could all reached 0.99, with the mean square error of 0.12 , 0.17 and 0.06 (MJ·kg-1)2, the forecasted average relative deviations were 1.2%, 1.23% and 0.69%. The results showed that the SVM regression method based on K-CV parameter-optimized could be used for quantitative analysis of coal heat value using LIBS technology, and could obtain higher analysis accuracy. At the same time, by comparing the quantitative analysis models using different spectral features, the quantification model of heat value by using the characteristic spectrum of non-metal plus metal elements, can effectively improve the accuracy of LIBS in the rapid detection of heat value in coal. This method can achieve accurate prediction of heat value in coal.