icPL-ANN近红外光谱分析方法在航空燃料中的应用研究  

Study on the application of icPL-ANN near infrared spectroscopy analysis technology in aviation fuel

作　　者：邢志娜[1] 王菊香[1] 瞿军[1] 刘洁[1] 

机构地区：[1]海军航空工程学院飞行器工程系,烟台264001

出　　处：《分析试验室》2016年第8期971-974,共4页Chinese Journal of Analysis Laboratory

摘　　要：以航空燃料的闪点预测为例,针对数据分布分散不连续,与光谱信息的线性关联偏弱的情况,提出一种将波段间隔组合与线性-人工神经网络(icPLANN)相结合的近红外光谱定量分析方法。该方法利用分段建模考核进行波段优选,最大程度地提取了有效信息,并结合PL-ANN方法建立了近红外光谱定量分析模型。最终把预测结果与间隔组合偏最小二乘法(icPLS)的实验结果进行了对比。结果表明,间隔组合PL-ANN模型的校正标准偏差(SEC)为0.75,预测标准偏差(SEP)为0.86,而间隔组合偏最小二乘法SEC为1.48,SEP为1.08,因此前一种方法的预测精度更高,预测决定系数(Rp2)能达到0.8971。可见,针对分散不连续数据与近红外光谱的复共线性影响预测模型准确度和稳定性的问题,间隔组合PL-ANN方法是一种有效的近红外光谱定量方法。The study of this article is illustrated with the determination of flash point of aviation fuel. A near infrared spectroscopy（ NIR） quantitative analysis method based on pureline artificial neural networks（ ic PL-ANN）with wavelength interval combination is presented to address the issues of discontinuous data and the weak line relation between spectrum and data. In this method,the available spectral information is picked up furthest by the means of wavelength intervals optimization with subsection modelling check. The result of the model built by ic PL-ANN is compared with that of the model built by interval combination partial least squares（ ic PLS）. Calibration standard error（ SEC） of the two models are 0. 75 and 1. 48 respectively,and prediction standard error（ SEP） of them are 0. 86 and 1. 08. The experimental result indicates that the new method has the better prediction accuracy. Its determination coefficient is up to 0. 8971. Therefore,ic PL-ANN is an effective NIR quantitative analysis method to reduce the modeling variables and improve the precision and robustness of the analysis model.

关 键 词：间隔组合线性神经网络(ic PL-ANN)法 近红外光谱(NIR) 波长优选 

分 类 号：TQ075[化学工程] O657.3[理学—分析化学]