【作者】 刘芳；

【导师】 王俊德；

【作者基本信息】 南京理工大学 ， 应用化学， 2003， 博士

【摘要】 本文主要对大气中有毒有机物傅里叶变换红外光谱的解析和室内扩散模型的建立进行了研究。在红外光谱解析方面，着重讨论了如何使用遗传算法对高度混叠的、难识别的谱图进行定量解析，对未知体系的定量识别，以及计算机联机谱图检索等方面的应用；利用小波的信号多尺度边缘特征提取，采用墨西哥帽函数小波对所获取的遥感傅里叶变换红外光谱信号进行连续小波变换，从而达到识别和定量解析遥感光谱的目的。在室内扩散模型的建立方面，选择了高斯模型作为本实验研究的理论基础。通过实验，研究了四种纯物质和一种混合物在室内空气中积分浓度随时间衰减的关系，获得数学模型。在高斯模型的基础上，进一步推导出高度、距离、时间与空间某一点浓度关系的三维模型，并与实测值进行了比较。 1．遗传算法 在化学计量学中，谱图识别特别是难以用常规方法识别的严重混叠谱图的解析，已成为广大化学工作者研究工作的重点。随着计算机容量增加和计算速度的大幅度的提高，遗传算法的研究开始蓬勃发展。由于该算法的整体搜索策略和优化计算时不依赖于梯度信息，所以它的应用范围极其广泛，尤其适合处理传统方法难以解决的高度复杂的非线性问题。遗传算法模拟了自然界生物优胜劣汰的进化过程，并逐渐逼近最优解。该法的实施包含了5个基本要素，即参数编码，初始群体的设定，适应度函数的设计，遗传操作设计和收敛的设定。本文着重讨论了如何将遗传算法应用于复杂的FTIR光谱谱图解析。文中包括对交叉概率的确定、变异率的确定、自适应变异步长技术的应用、适应度尺度的确定、收敛条件的讨论等方面的研究工作；以及对高度混叠的、难识别的傅里叶变换红外谱图的定量解析；对未知体系的识别；计算机联机谱图检索等方面的应用。 研究结果表明，遗传算法能够成功地用于FTIR混叠光谱定量解析。其优点如下：不需要确切的解集空间；具有全球搜索特性；结果的相对误差小且精确；有着良好的非线性性。遗传算法也是有关谱图的谱库检索的一个很好的工具。具有过程简单，结果比较准确的特点，很适合低浓度，峰形不明显的谱图检索。 此外，遗传算法在标准谱图数据库的帮助下，能够成功地定性和定量识别未知FTIR光谱体系。与ANN和PLS相比，减少了对数据的预处理的工作量。特别是对于难于用常规方法解析的谱图，使用遗传算法可以获得很好的效果。该算法过程简单，只需合理设置与求解问题有关的目标函数，经过交叉、变异等遗传算子操作，就可博士论文大气中有毒有机物Fl，IR光谱解析技术及其室内扩散模型建立的研究得到比较精确的结果。2.小波变换 对于一个遥感FTIR光谱仪，有时受到周围气象因素和背景变化的影响，实时监测所得到的谱图往往难以辨认。如何从复杂的谱图中提取有用信号，便是个值得探讨的问题。本文就是利用小波的信号的多尺度边缘特征提取，对遥感FTIR光谱数据进行小波分解，提取有用信息，并将其应用于遥感FTIR光谱图的定性和定量解析，从而达到识别和解析光谱的目的。此外，还研究了FTIR谱图的基线漂移与小波变换的关系。 研究结果表明，小波分析能够在一定程度上对FTIR光谱图进行信号的特征提取，特别是对于纯谱图的信号或是信号比较突出的谱图定位准确。而且，在分解过程中，能够对谱图中的噪声进行一定程度的滤除，突出有用信号。 此外，可以通过对标准谱图的小波分解的研究，获得小波分解系数与相应物质浓度之间的关系，从而进行谱图的定量解析工作;可以通过某物质的不同特征吸收峰处的浓度预测值之间的误差，来判断该物质是否含有未知组分或干扰。同时，标准谱图的小波变换信息也可用于对遥感FTIR谱图的解析。小波变换技术的另外一个优点是，对于基线漂移的谱图，它的小波变换系数与修正基线后的谱图一致。这样，对于基线漂移的谱图，其光谱数据无需处理，可直接进行小波变换。3.室内扩散模型建立 近年来，除了对大范围的环境大气质量变化得到广泛的研究以外，对室内空气质量研究也开始活跃。主要是因为人类在室内的活动时间较长，室内大气污染对人类健康关系密切。建筑物内部的表面性质，装饰装修材料所散发的各种有机物质，个人爱好、行为以及文化修养等将对室内空气质量产生很大的影响。 由于不同有机物质其性质的不同，在室内扩散的情形也是千差万别。因此，到目前为止，还没有适合于一切情况的统一的模式。鉴于现在的研究状况，本文采用了由实践到理论的研究方法。研究了单‘一组分物质在不同源高下，在室内空气中一定测定距离的积分浓度随时间衰减的关系，建立了相应的数学模型。在上述模型基础上，以高斯模型为理论基础，进一步推导了几种物质在室内比较理想条件下的逸散的三维模型，通过模型预测出任意时间，任意一点的某物质在空气中的浓度。检测的结果显示，这几种物质中，预测最大相对误差为46.3%，最小为3.5%。 另外，从所得到的结果来看，有机物质在室内的扩散还是有一定规律可寻的。如点源释放某物质时，可根据到达最大积分浓度及其所需的时间，预?更多还原

【Abstract】 The paper mainly focuses on the analysis of FTIR spectra of toxic organic substance in the atmosphere and the establishment of diffusion models in the indoor air. Genetic Algorithm (GA) is used to quantitatively analyze strongly overlapped and undistinguishable spectra. It is also used to search pure spectra in spectra database by computer. Wavelet Transform (WT) technology of multiresolution feature extraction is studied to identify and quantitatively analyze remote sensing FTIR spectra by Mexican hat function. The signal is performed by continuous wavelet transform (CWT). Gaussian model is selected as a basic one to establish diffusion models in the indoor air. The attenuation relationships between integral concentrations and time of four pure subjects and one mixture are studied and modeled by mathematics. Three-dimensional models, which contain height, distance, time and concentration of a certain space point, are educed. The predicted values obtained from three-dimensional models are also compared with the real values.1. Genetic Algotithm (GA)How to identify strongly overlapped spectra has been a keystone in the chemometrics. With the rapid increase of the computer capability and calculating speed, the research on GA enters into a new course of development. It is widely used in many fields because it doesn’ tdepend on grads information. Especially, it can deal with very complicated non-linear problem, which can not been solved by traditional method. GA simulates the evolution process of" the superior winning and the bad being eliminated" in nature and approaches the excellent results gradually. Five basic factors, i. e. parameter coding, determination of initial population, design of fitness function, design of heredity operation and the determination of convergence, are included in GA. Application of GA in the analysis of FTIR spectra is mainly discussed in the paper. The determination of the probability of crossover, probability of mutation, application of adaptive mutation, determination of fitness scale, discussion of convergent condition, quantitative analysis of mixed and undistinguishable FTIR spectra, identification of unknown system and spectrasearch on line are studied.The results indicate that GA can quantificationally analyze FTIR overlapped spectra successfully. It has several virtues, such as stochastic solution space; its function of global search; small relative errors and well non-linear characteristic. GA is also used as a good tool to search spectra. The search process is simple and the results are exact. It is suitable for spectra search, which have low concentrations and unsharp absorption peaks.Moreover, GA can identify quantificationally unknown FTIR spectra with the help of standard spectra database. Compared with ANN and PLS, it needn’ t pretreat spectra data and can obtain a good effect for some spectra, which can not be analyzed by normal methods. Its calculating process is also simple. The accurate results can be obtained when the objective function is set up rationally and the data are properly performed by crossover and mutation operator.2. Wavelet Transform (WT)The spectra obtained from real-time detection in Remote Sensing FTIR spectrometer sometimes can not be identified because of the influence of the weather factors and the change of background. How to extract useful information from the complicated spectra is an attractive problem. WT technique is applied here to decompose FTIR spectra data, extract useful information and analyze FTIR spectra quantificationally for the purpose of spectra identification and analysis. Moreover, the relationship between WT and the baseline excursion of FTIR spectra is studied.The results indicate that WT can extract information from FTIR spectra to certain degree. Especially, it can determine the position accurately for pure or relatively outstanding spectra. At the same time, the noise in the spectra can be eliminated and the useful information is magnified.Furthermore, the relationship between WT更多还原