近红外定性分析模型的稳健性与适应性分析  

Analysis of the Stability and Adaptability of Near Infrared Spectra Qualitative Analysis Model

作　　者：曹吾 李卫军[1] 王平[2] 张丽萍[1] 

机构地区：[1]中国科学院半导体研究所、高速电路与神经网络实验室,北京100083 [2]中国石油大学(华东)、信息与控制工程学院,山东青岛266580

出　　处：《光谱学与光谱分析》2014年第6期1506-1511,共6页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(90920013)资助

摘　　要：研究了近红外定性分析中模型的稳健性与适应性问题。讨论了单独建模延长建模周期对模型性能的影响,结果显示这种方法可以显著提高模型的稳健性,同时对模型的台间适应性也有所提升,但提升作用有限;研究了不同仪器联合建模的方法对模型性能的影响,该方法不仅可以显著提高模型的适应性,而且可以有效提高模型的稳健性。对比单独建模,可以缩短建模时间,减少模型建立的工作量;延长模型的适用期限,提高建模效率。结果表明,对模型适应性的测试,单独建模其正确识别率较低,不能满足应用的要求,而多台仪器联合建模可以达到90%以上,提升效果明显;对模型稳健性的测试,联合建模也能得到相比单独建模更好的模型识别效果,具有很好的应用价值。The stability and adaptability of model of near infrared spectra qualitative analysis were studied.Method of separate modeling can significantly improve the stability and adaptability of model,but its ability of improving adaptability of model is limited.Method of joint modeling can not only improve the adaptability of the model,but also the stability of model,at the same time,compared to separate modeling,the method can shorten the modeling time,reduce the modeling workload;extend the term of validity of model,and improve the modeling efficiency.The experiment of model adaptability shows that,the correct recognition rate of separate modeling method is relatively low,which can not meet the requirements of application,and joint modeling method can reach the correct recognition rate of 90%,and significantly enhances the recognition effect.The experiment of model stability shows that,the identification results of model by joint modeling are better than the model by separate model-ing,and has good application value.

关 键 词：近红外光谱 定性分析 模型期限 模型传递 建模周期 联合建模 

分 类 号：G202[文化科学—传播学] S123[农业科学—农业基础科学]