【作者】 王春艳；

【导师】 郑荣儿；

【作者基本信息】 中国海洋大学 ， 海洋信息探测与处理， 2010， 博士

【摘要】 海洋溢油事故的频发和其对海洋环境安全和人类健康的严重危害,将海洋溢油的研究提到了全球环境问题的焦点之一。溢油事故发生后,必须及时准确地进行油指纹分析鉴别,确定责任归属,为污染清除费用的索赔提供依据,这对有效地防治船舶运输、船上石油开发造成的油污染具有重要意义。因此,建立一套实时、经济、易于推广的海洋溢油样品多环芳烃(PAHs)分析技术,对中国这样一个环境压力日趋严重的世界上最大的发展中国家具有重要的实用价值。荧光光谱技术具有灵敏度高、分析结果快速、受风化影响小等优点,一直作为国际海事组织(IMO)推荐的主要标准的化学分析仪器之一,但光谱重叠严重,对于相近油源原油样品的鉴别能力有限,溢油指纹检测仅限于初期的普查阶段,未能成长为独立有效的检测手段。作为成熟的光谱技术,荧光技术近年来的发展趋势主要集中在(i)针对荧光鉴别能力有限的问题,改进光谱获取技术,增加信息维度,丰富荧光信息。(ii)针对多组分混合体系荧光光谱重叠严重的问题,利用先进的数据挖掘手段提取特征量,进行光谱分析及识别。论文针对原油相关样品,从荧光光谱技术的发展及数据挖掘手段的应用两个方面做了较为详细的综述。同时论文的主体工作也围绕这个两个方面进行了探索,以期提高荧光指纹鉴别技术的识别率。论文的第三章针对原油相关样品荧光光谱的光谱特征,提出了引入浓度作为辅助参量以增加荧光光谱信息量的方法,结合单质芳烃组分的荧光光谱随浓度变化的实验研究分析引入浓度参量的必要性和有效性。针对原油相关样品浓度参量荧光光谱的特点,就光谱技术的选择,参数的选择,试剂的选取进行了详细的实验分析和讨论。考虑到单一浓度不能反映原油相关样品芳烃组成比率的不同,引入浓度一维,全面反映不同环数的多环芳烃及其荧光特性,同时利用同步荧光光谱可通过单次测量反映三维光谱的主要信息。二者的结合构成的浓度同步荧光光谱矩阵(CSMF)全面地反映了原油相关样品芳烃组分荧光信息,为数据挖掘提供充足的组分信息。第四章基于浓度同步荧光光谱矩阵(CSMF)对不同层次原油相关样品进行光谱数据的采集。考虑到海洋溢油复杂条件对浓度参量荧光光谱技术可能造成的影响,在多种外扰条件进行了综合实验研究,获取了不同类别不同油田36个原油相关样品的光谱。从光谱角度全面考察了荧光溢油鉴别方法的有效性和适用范围。有效的特征提取是模式识别成功率的关键,第五章利用不同的特征提取方法对不同层次的原油相关样品进行了特征提取,并针对其特征提取量特点对方法的有效性进行了详细讨论。结果表明：主成分分析方法(PCA)的主成分载荷图可以很好的反映各个原油相关样品在油源上相近程度；而偏最小二乘法PLS的主成分提取方式在相近油源的原油相关样品的特征分析上要优于PCA；二维Gabor小波变换能够捕捉对应于空间位置、空间频率及方向选择性的局部结构信息,对风化条件下的相近油源样品的CSMF光谱实现了鲁棒性高的最优特征提取,为分类识别奠定了良好基础。第六章是在第五章浓度同步荧光光谱矩阵(CSFM)详细的特征分析的基础上,针对原油相关样品样本进行最终的分类识别。根据海洋溢油现场的要求,通过“浓度层析局部匹配方法”与“相对特征提取及模式识别方法”的结合,建立起一套基于原油样品浓度参量荧光光谱的溢油指纹鉴别技术。首先利用浓度层析光谱局部匹配方法分别对相近油源的原油相关样品鉴别和引入外扰原油相关样品集进行了测试。通过对不同参数选择的测试结果分析,对该方法参数的选定进行了讨论,确定了参数的最佳选择范围。结果表明,该方法不需要大量对可疑油样提取训练样本,只需单次测量可疑油样的浓度同步荧光光谱矩阵(CSFM),同时对肇事油样,只需采集6-8个浓度系列的CSFM光谱,十分符合快速、简便、易操作的溢油实时监测要求,并可同时对溢油在一定体积中的含量进行定量。但风化等外扰的引入也会使相近油种之间鉴别产生错误判断,准确率降低。当外扰较大的情况下,可以辅助其他指纹鉴别方法实现浓度定量。作为本论文的各类数据分析方法的总结,将PCA、PLS以及二维Gabor小波提取的特征量,分别与ANN和SVM结合进行原油相关样品的分类识别。在特征提取的选择上,Gabor要优于PCA和PLS,而针对模式识别,由于该油种鉴别技术是属于小样本分类识别,所以ANN的结果不稳定,SVM的分类效果较为理想。交叉检验的结果表明：针对引入外扰情况下的相近油源的溢油样品集,特征提取和分类器的选择对识别结果的影响较大,其中Gabor_SVM的识别准确率是最高的,可达到92%。同时由于两类分类区分的结果远远高于多类分类识别的结果,可利用多类识别的方法,逐步缩减嫌疑样本,最后实现原油相关样品的准确识别。作为论文的最后一部分,第七章在对论文工作进行总结的基础上,从方法改进和实用化两个角度提出了进一步努力的方向：实用化技术方案的充实完备、样品池的进一步改进、新的数据挖掘方法的引入、仪器的初步构想以及将本论文中实现的方法向其它领域扩展的设想。更多还原

【Abstract】 Fingerprinting Technique of Oil Species Based on Concentration Parameter Fluorescence SpectraThere has been a growing concern in recent years about increasing occurrence of spilled oils to the environment and proven toxic potential of these pollutants on human health and wildlife. The existence of these harmful substances in the environment has disrupted the natural cycles and processes and caused great economic loss to nations rich or poor. Precisely determining the sources of spilled oils can provide scientific evidence for the investigation and handling of spilled oils accidents.The development and implement of a method that is efficient, economic and easy to use routinely could offer decision-makers and model developers preliminary information of spilled oils in a short period, while the complicated approaches could provide more detailed information afterwards.The oil-bearing samples contain traces of polycyclic aromatic hydrocarbons (PAHs) that are highly fluorescent. Fluorescence-based techniques feature high sensitivity, good diagnostic potential, relatively simple instrumentation and suitability for portable instrumentation. Unfortunately, the chemical and physical complexities of crude oils and petroleum products lead to broad spectra without fine structures.Two main areas of interest distinguished themselves in fluorescence techniques over the last two decades:Multi-dimensional Fluorescence techniques to obtain more fluorescence information of multi-fluorehore mixtures, and the applications of chemometrics in spectroscopic study.The thesis begins with an overview of spilled oils fingerprinting technique and the development of Enviromental Forensics, followed by a detailed review of relevant studies, including (i) application of fluorescence techniques to petroleum-related samples and the concentration-depended fluorescence studies of PAHs and (ii) application of chemometrics to enviromental analysis of PAHs. In this thesis, two recent versions of this technique, Multi-dimensional Fluorescence techniques and data mining methods, have been applied to the analysis of spilled oils samples to improve identification accuracy.In Chapter 3, concentration-dependent fluorescence study of single PAH molecules and petroleum-related samples were presented first. Based on a detailed discussion on different spectrum approaches and extractants, the author described a novel method she developed for species identification of petroleum-related samples using the concentration auxiliary parameter synchronous fluorescence technique. By introducing concentration value as a parameter, a new Concentration-Synchronous-Matrix-Fluorescence (CSMF) spectrum was formed with a series of synchronous fluorescence spectra (SFS) at different levels of concentration. It was observed that the SFS varied with concentration level and the profiles of CSMF spectra changed from species to species. Therefore, CSMF spectra can be used for species identification.A detailed experimental investigation on different levels of petroleum-related samples is given in Chapter 4, along with the consideration of various disturbances, such as weathering, adulation of seawater, change of light source intensity, mixture of different oils, among others. The CSMF spectra of 36 petroleum-related samples from different oil-spill types had been obtained and three data sets were chosen to assess the feasibility and performance of the feature extraction and pattern recognition methods used in this thesis work.The author’s main work is to perform data mining of the CSFM spectra, which is described in Chapters 5 and 6. Effective feature extraction is the key to accurate pattern recognition. In Chapter 5, principal components analysis (PCA) and partial least square analysis (PLS) are used to extract the main orthogonal contributions, which explain most of the variance of the spectra measurement matrix. The results show that the PCA can divide the samples to different oil types in the principal components space, while PLS can give a better classification of the closely-related source of petroleum-related samples due to its ability to find the multi-dimensional direction in the measurement matrix space that explains the maximum direction in the response vector space. The CSMF images transformed by Gabor wavelet exhibit strong characteristics of spatial locality, and scale and orientation selectivity, and Gabor is shown to be the best feature extraction method to the pattern recognition.The work presented in Chapter 6 was carried out to measure the effectiveness, computing speed, and accuracy of the classification methods used in this thesis. The partial surface fitting to CSFM with interpolation was introduced first. With surface fitting, CSMF spectra of the closely-related source crude oil samples were successfully discriminated, and the initial concentration of the test samples was also obtained. Large disturbances, however, result in low accuracy of discrimination.The feature extraction methods, such as PCA, PLS and Gabor wavelet, combining with the pattern recognition methods, such as artificial neural network (ANN) and supported vector machine (SVM), were used to identify the CSFMs of the data sets introduced in Chapter 4. An ideal result of closely-related oil source samples with the 92% of the correct rate of oil species recognition is achieved by combining Gabor wavelet with SVM. The obtained results suggest that the newly-developed method may become a more specifically applicable means in spilled oils identification.In the last chapter of the thesis, a thorough discussion of the thesis work is given. In addition, suggestions for future work are provided, including direction of data mining, concept of new instrumentation design, and several additional experiments, which should lead to a better understanding of the mechanisms involved in concentration-depended fluorescence spectra.更多还原