基于RBF神经网络的较低浓度下同步荧光光谱的溢油鉴别

针对海面溢油样品的含量难以确定,同时考虑到海水掺杂及风化等问题的影响,提出了在较低非线性浓度范围内采集溢油嫌疑样品的同步荧光光谱,获取其训练样本集,利用主成分分析法(Principal com-ponent analysis,PCA)提取其特征光谱,结合径向基函数(Radial basis function,RBF)神经网络对肇事样本和嫌疑样本进行模式识别的方法。通过对相近油源原油样品分类识别研究表明:该方法仅需单次对肇事样本同步光谱测量,再借助数据分析,就可以很好区分相近油源溢油样品,外扰对识别率影响也不大。RBF神经网络算法识别率在92%左右。该结论对海洋环境中溢油的实时检测及油指纹数据信息库的建立有重要意义。

Identification of spill oil species based on low concentration synchronous fluorescence spectra and RBF neural network

In this paper, a new method was developed to differentiate the spill oil samples. The synchronous fluorescence spectra in the lower nonlinear concentration range of 10(-2) - 10(-1) g x L(-1) were collected to get training data base. Radial basis function artificial neural network (RBF-ANN) was used to identify the samples sets, along with principal component analysis (PCA) as the feature extraction method. The recognition rate of the closely-related oil source samples is 92%. All the results demonstrated that the proposed method could identify the crude oil samples effectively by just one synchronous spectrum of the spill oil sample. The method was supposed to be very suitable to the real-time spill oil identification, and can also be easily applied to the oil logging and the analysis of other multi-PAHs or multi-fluorescent mixtures.