论文部分内容阅读
作为煤质评价的重要指标之一,热值的快速、准确测量对电厂燃煤锅炉的优化燃烧和经济运行至关重要。采用激光诱导击穿光谱(LIBS)技术结合BP神经网络定量分析模型和聚类分析,以35个煤粉样品作为研究对象进行热值的定量分析。基体效应对LIBS光谱数据的显著影响,针对基于某类煤粉样品所建立的定标曲线不能直接用于不同煤种的定量分析,采用K-means聚类方法根据热值、灰分、挥发分把样品分为三类对训练集和预测集样品进行优化选择。通过谱线强度和热值变量相关性分析,同时考虑特征谱线的物理意义,最终提取12条元素谱线的峰值强度作为输入参数,建立BP神经网络模型对燃煤热值进行预测。定标结果表明,建立的神经网络模型具有良好的定量分析能力,定标曲线拟合度R2为0.996,热值预测值的相对误差低于3.42%,多次重复测量的相对标准偏差在4.23%以内。对聚类分析中3类样品具有不同的预测能力,采用峰值强度作为输入参数时,能够在一定程度上减弱试验参数波动和基体效应造成的影响。定量分析结果的重复性和准确性可以通过对不同类别的煤种分别建立BP神经网络模型来进一步改善。LIBS技术结合BP神经网络可以对煤粉热值进行定量分析,在现场在线/快速检测领域具有很好的应用价值和潜力。
As one of the important indexes of coal quality evaluation, the rapid and accurate measurement of calorific value is crucial to the optimal combustion and economic operation of power plant coal-fired boilers. The laser induced breakdown spectroscopy (LIBS) combined with BP neural network quantitative analysis model and cluster analysis were used to quantitatively analyze the calorific value of 35 coal samples. The effect of substrate effect on LIBS spectral data is significant. For the calibration curve established based on some pulverized coal samples, it can not be directly used for quantitative analysis of different coal types. The K-means clustering method is used to calculate the temperature, The samples are divided into three categories to optimize the selection of training set and prediction set samples. Through the correlation analysis between spectral intensity and calorific value, considering the physical meaning of characteristic spectral lines, the peak intensities of 12 elemental spectra were finally extracted as input parameters, and a BP neural network model was established to predict the calorific value of coal. The calibration results show that the neural network model has a good quantitative analysis ability, the calibration curve fitting R2 is 0.996, the relative error of the predicted value of calorific value is less than 3.42%, the relative standard deviation of repeated measurement is 4.23% Within. Three kinds of samples in the cluster analysis have different predictive ability. When using the peak intensity as the input parameter, the influence of the fluctuation of the test parameters and the matrix effect can be weakened to a certain extent. The repeatability and accuracy of quantitative analysis results can be further improved by establishing BP neural network models for different types of coal. LIBS technology combined with BP neural network can quantitatively analyze the calorific value of pulverized coal and has good application value and potential in on-line / rapid field detection.