【作者】 田广军；

【导师】 史锦珊；

【作者基本信息】 燕山大学 ， 测试计量技术及仪器， 2005， 博士

【摘要】 各种矿物油是水环境的主要污染物。快速准确地测定水中矿物油的种类及含量,正确判断油污染的性质和来源,对于实施水环境监测管理及污染防治具有重要意义。本文在对国内外关于水环境油污染荧光分析技术的现状及发展趋势进行分析和综合的基础上,针对矿物油的种类鉴别和浓度测定及其在水环境监测中的应用,提出了基于三维荧光谱参数化的矿物油神经网络模式识别方法,并进行了实验研究。 首先以荧光原理及测量技术为基础,组成荧光谱测量系统,对柴油、煤油、机油、和原油等多种矿物油进行了激发-发射荧光光谱扫描测量和实验研究工作,获得了充实的矿物油荧光谱第一手资料;对大量的原始数据进行了处理,完成了矿物油三维荧光谱的重构和分析;研究了矿物油荧光谱最大值与浓度的定量关系,给出了荧光强度及其参考光强随着浓度变化的关系曲线。 研究了基于表观统计特征的三维荧光谱参数化方法。提取和计算了多种实际油样三维荧光谱的表观统计特征参数,并通过对典型样本聚类的方法,讨论了溶剂等因素对三维荧光谱的影响(选用水溶剂作为本底标准)。应用结果表明了矿物油三维荧光谱参数化的有效性及模式识别的可行性,同时也表明了表观特征参数的局限性。 为了寻找更为精细的特征参数,组成具有深层物理意义的特征谱,有效鉴别荧光谱有相互重叠的各种油类或多组分污染油,对激发-发射矩阵进行了数据挖掘。通过系统聚类分析、比较,选择了奇异值特征谱作为矿物油种类鉴别信息。 利用激发光的二级色散光作为荧光的参比标准,导出了与光源稳定性、系统增益近似无关的矿物油浓度信息,提高了浓度测量的准确性。 将矿物油的种类和浓度两种信息相结合,组成便于模式识别的油类三维荧光谱的综合特征谱序列,为解决污染油定性和定量分析提拱了基础。 设计了定性和定量双重处理的前向神经网络。用鉴别网络确定种类后,反馈到浓度网络的输入端,和相对荧光强度一起预测相应种类的浓度输出。通过对矿物油的综合特征谱进行模式识别,同时实现了水中污染矿物油的种类鉴别和浓度测量。更多还原

【Abstract】 Mineral oils are the principal contaminations in water environment. Identifying the species of mineral oil in water and measuring its concentration are needed to determine the contamination oil source and evaluating pollution fast and correctly. It has important significance for water environment protection under control. Based on comprehensive analysis and synthesis of the present conditions, as well as its tendency of contamination oil fluorescent analysis techniques, this paper is aimed at identification and concentrate measurement of contamination mineral oil, as well as its application in water environment detection or monitoring. Based on experiment and parameterization of three-dimensional fluorescence spectra, oil pattern recognition using artificial neural network is presented. Based on fluorescence principle and its measurement technique, many kinds of mineral oils such as diesel, kerosene, engine oil and crude oils were detected and researched in excitation–emission fluorescence scanning experiments. Firsthand materials of large amount of fluorescence data were obtained, based on which many three-dimensional spectra were reconstructed visually, examined and analyzed by programmed data processing. Quantitative relationship between the maximal fluorescence value and concentration were discussed. The intensities curves of fluorescence versus concentration were given with reference lines companioned Parameterization based on apparent features of 3D fluorescence spectra was studied, and apparent statistic parameters of various oils were extracted and calculated. The effect of some factors such as solvents on three-dimensional spectra was visually discussed by parameter clustering for typical samples (distilled or pure water was selected as the background), and by this way, the effectiveness of parameterization was demonstrated. However the limitation of the parameters based on appearance manifested. To find sophisticated feature parameters forming profound feature vector with more physical meaning, which should be able to identify various or complex contamination oils with similar or overlapping fluorescence spectra, deeper data mining was carried on. As the result of systematic clustering and validity comparing, the singular values of the excitation-emission matrix (EEM) were selected as the identification information. The concentration information is get independent to the stability of excitation light as well as the system gain by making use of the second-order dispersion light working as a surveyor’s pole. By this way, the accuracy of concentration measurement can be improved. By associating species and concentration information, associated features of three-dimensional fluorescence spectra of oil have been made up of, providing foundation for solving the problem of both qualitative and quantitative analysis of contamination oil in water. A double neural network has been designed to implement qualitative and quantitative processing together. The result of identification is fed back to the input of concentration net, wherein the identification result and relative fluorescence intensity are fused to predict oil concentration of corresponding species. At last, with associated features as the pattern input, various samples of contamination oils in water have been identified correctly and measured with an acceptable accuracy.更多还原