Abstract: Based on the analysis of Infrared Spectroscopy curve obtained from different traits asphalts, we used the principal component analysis to make a cluster analysis on the spectrum data, and established the discrimination model for different asphalt with different traits by using the extracted principal component as the input value of the BP neural network. We took the first three principal components of the model as the input variables of the neural network, which accelerated neural network training of and improved the precision of the model. We randomly selected total 110 data samples from asphalt samples from each kind of trait to form a training set, the remaining 30 samples formed the prediction set. Then we established the training model and used the samples of prediction set to verify the model, and determined the standard deviation as ± 0.01. The result showed that only one unknown sample beyond the deviation range and the correct rate of this method was 96.7%. The satisfactory result suggested that the Infrared Spectroscopy curve had excellent effect on classification and identification, and provided a new method for the rapid identification of different traits of asphalts.

Key words: Infrared Spectroscopy, asphalt, aged asphalt, principal component analysis, BP neural network

摘要： 通过对不同性状沥青获取的可见红外光谱曲线的分析，采用主成分分析方法对光谱数据进行聚类分析，并将提取的主成分作为BP神经网络的输入值建立了不同性状沥青判别模型。该模型将前3个主成分作为神经网络的输入变量，加速了神经网络的学习速度，提高了模型的预测精度。随机选取每个性状的22个沥青样本共110个样本组成训练集，剩余的30个样本组成预测集，建立训练模型，并用预测集样本对其进行验证，将判定的偏差标准定为±0.01，结果表明只有1个未知样本超出偏差范围，该方法的判定正确率为96.7% ，获得了满意的结果。说明采用可见红外光谱判别具有很好的分类和鉴别作用，为不同性状沥青的快速判别提供了一种新方法。

关键词: 红外光谱, 道路石油沥青, 老化沥青, 主成分分析, BP神经网络