基于偏最小二乘的鱼类超微弱发光建模方法研究  

Modeling method research of ultra-weak photon emission of fish based on PLS

作　　者：叶廷东[1] 洪晓斌[2] 

机构地区：[1]广东轻工职业技术学院计算机工程系,广州510300 [2]华南理工大学机械与汽车工程学院,广州510640

出　　处：《计算机应用研究》2013年第10期2962-2964,3006,共4页Application Research of Computers

基　　金：广东省自然科学基金资助项目(S2012040007521);中国博士后基金资助项目(20070420779);广东轻工职业技术学院基金资助项目(KJ201215)

摘　　要：针对鱼类超微弱发光信息建模中存在的信息降维、多重相关性和非线性问题,提出了一种基于变量筛选与双层多项式非线性偏最小二乘的多维信息建模方法。该方法用偏最小二乘方法进行变量筛选,然后采用双层多项式非线性偏最小二乘方法实现对多维信息的非线性建模。检测应用的结果表明,该方法与偏最小二乘方法相比,可以在少用拟合自变量的情况下,预测准确度分别提高了56.2%和24.7%,实现了对多维信息的有效降维与非线性建模,以及鱼类超微弱发光的准确检测。This paper aimed at problems of dimension reduction, multiple correlation and nonlinearity existing in the informa- tion modeling of ultra-weak photon emission of fish ; it advanced a multi-dimensional information modeling method based on variable selection and double-layer polynomial NPLS （nonlinear partial least squares）. Firstly, the method used PLS （partial least squares）method to select variable, then used double-layer polynomial NPLS to realize nonlinear modeling of multi-infor- mation. The application result shows that the prediction accuracy of the advanced method has improved 56. 2% ,24. 7% re- spectively, and it uses less fitting variable compared with PLS, it realizes valid dimension reduction, nonlinear modeling of multi-information, and accurate detection of ultra-weak photon emission of fish.

关 键 词：偏最小二乘 变量筛选 建模 多项式 超微弱发光 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TP301.6[自动化与计算机技术—控制科学与工程]