采用恒波长同步荧光光谱法检测分析8种燃料油、 7种中东原油、 14种非中东原油的荧光特征, 结合离散小波变换以及Fisher判别法建立海上溢油油种鉴别的模型。 29种油样风化前后均在波长(280±2), (302±2), (332±2)和(380±2) nm处有典型的荧光峰, 但在(380±2) nm处风化油样的荧光强度的离散度过大, 该波长不适于油种鉴别。 在db7小波基函数下对29种原始油样荧光谱图进行6层分解, 提取d3细节系数特征, 确定波长(255±2), (280±2), (302±2), (332±2)和(354±2) nm处的小波系数并用于Fisher判别模型建立。 29种油样在(280±2) nm处均有极值点, 燃料油小波系数位于44.06±5.62之间, 原油位于22.47±5.12之间, 此波长处的小波系数可区分燃料油与原油。 建立的Fisher判别模型不仅能区分燃料油和原油还能进一步区分中东原油, Wilks’s lambda分布所对应的P值分别为0和0.02, 表明模型是可行的。 模型验证结果显示, 对风化后的建模油样的鉴别正确率达到96.6%, 对非建模23种油样鉴别正确率达到95.7%。 由于建模油样风化前后的修正余弦相似度为0.91~0.98, 因而以未风化油样建立的油种鉴别模型同样适用于风化后油样的辨别。

The fluorescence characteristics of 8 kinds of fuel oil, 7 kinds of Middle East crude oil and 14 kinds of non-Middle East crude oil were analyzed by a constant-wavelength synchronous fluorescence spectrometry. Discrete wavelet transform and Fisher discriminant were combined to establish a model for the identification of marine oil spills. Twenty-nine kinds of oil before and after weathering had typical fluorescence peaks at the wavelength of (280±2), (302±2), (332±2) and (380±2) nm, but the dispersion degree of fluorescence intensity of weathered oil at (380±2) nm was too high, which was not suitable for the identification of oil species. The db7 wavelet basis function was used to resolve 6 levels for fluorescence spectra of 29 kinds of original oil samples, and the d3 detailed coefficient characteristics were extracted. The wavelet coefficients corresponding to (255±2), (280±2), (302±2), (332±2) and (354±2) nm were determined which were used to establish the Fisher discriminant model. All of oil samples all had extreme points at (280±2) nm, the wavelet coefficients of marine fuels were between 44.06±5.62, and that of crude oils were between 22.47±5.12. These two wavelet coefficients could be used to distinguish the marine fuel and crude oil. The established Fisher discriminant model distinguished not only marine fuel and crude oil but also further Middle East crude oil. The P-values corresponding to Wilks’s lambda distribution were 0 and 0.02, respectively, which indicated the model was feasible. The validated results of the model showed that the identification accuracy reached 96.6% for modeling oils after weathering and 95.7% for 23 kinds of non-modeling oils. Since adjusted cosine similarity of modeling oil before and after weathering was ranged from 0.91 to 0.98, the identification model established by the original oils can also be used for the identification of weathered oil species.