【作者】 陈国松；

【导师】 时钧； 陆小华；

【作者基本信息】 南京工业大学 ， 化学工程， 2003， 博士

【摘要】 分光光度法具有灵敏、准确、快速、操作简便和仪器廉价的特点，应用非常广泛。但由于显色反应影响因素多，非特征性强，资料以描述型和经验型为主，有关新显色剂和新显色体系的文献又不断涌现，使非专家用户在方法的取舍上无所适从，甚至专家级用户也颇感为难。常规的专家系统研制思路似乎也不适于该领域问题的解决。因此，尽管预期的影响和意义重大，但分光光度分析专家系统目前仍是一个空白，显示出该项研究的工作量和难度很大。综合考虑分光光度分析方法的特点和专家系统的特点，提出了本文研究工作的思路： ①根据对分光光度分析的理解、前期工作的经验和用户需求的分析，总结分光光度分析的基本规律和关键影响因素，拓展分光光度法的功能，提高分光光度分析的研究水平； ②通过实验测量和文献检索，收集各类分光光度分析体系的“知识”，构建适宜表达这类“知识”的数据库； ③提出量化的分光光度分析方法筛选规则； ④针对分光光度分析最大的弱点——选择性差，根据文献和前期的研究工作，选取几种最具实用价值的化学计量学算法进行详细研究并做出改进，赋予分光光度法多元定量功能； ⑤建立一套适用于非专家和专家用户的实用、高效和开放的分光光度分析专家系统。 金属元素分光光度分析专家系统的主要使用对象定位于以常规分光光度法为手段对各类试样中金属元素的浓度或含量进行定量测定的非专家用户。因此需要将分光光度分析方法的经验规律提炼和上升到能用量化的数学语言表达的水平。本文强化了专家系统在显色剂选择和多元定量计算方面的功能，并在试样处理方面做了基础性的工作。以通用办公软件Access2000建立数据库，以Visual Basic 6.0为主要开发工具，用SQL7.0对数据库进行操作。在用户引导等人性化设计方面作了许多工作，突出专家系统的多功能、全面性、易用性和开放性，力争使非专家用户不经学习可以直接使用，甚至能进行修改、添加等维护工作。软件开发难度很大。 收集了202种显色剂和1202个显色体系的资料，并对其进行加工整理，将有效信息录入数据库。通过对显色反应的影响因素进行全面的比较和研究，在兼顾灵敏度、选择性的前提下，提出以摩尔吸光系数ε、对比度△λ和显色反应酸度为权重因子。选择具有代表性的显色体系，通过与专家经验对照，确定了各因摘要子的权值并进行了验证，表明系统推荐的结果与专家经验一致，在这方面基本达到了专家的水平。并且可以给出相关的详细资料和吸收曲线等一般手册所没有的信息。对多元测定，提出了以“条件数”为判据，对吸收谱线的重叠度进行评价，比专家的直观判断更迅速、更合理。并以多元稀土离子一偶氮氯麟m显色体系为例对上述情况进行了比较和讨论。 考虑到试样溶解的复杂性，以国家标准方法作为这一部分研究工作的前期基础性资料。对《中华人民共和国国家标准目录总汇(2002)}}进行了手工完全检索，从全部1606个分类中，收集了47个分类中关于金属离子分光光度测定的373个标准。将其制成PDF文件，并提取关键词录入Access数据库中，制作了用户引导界面，使用户在引导下可以查寻到相关的标准作为试样溶解方法的参考。 对于多元分光光度定量测定，选取了通用模拟退火算法(GSA)、偏最小二乘法(PLS)、BP人工神经网络算法(BP~ANN)和迭代目标转换因子分析法(ITTFA)这四种最具实用价值的化学计量学算法进行了详细的研究。提出了“回火过程”和“变步长搜索”的方式对模拟退火算法进行了改进，形成了“模拟退火一回火算法”，在保持原算法能收敛于全局最优的特点同时，加速了算法跳出局部最优的过程。提出“误差中位数反向传播”的思想，形成了“稳健BP人工神经网络算法”，有效地限制了随机误差对计算结果的不利影响，提高了算法的稳定性，并用蒙特卡罗方法进行了验证。将BP算法与偏最小二乘法进行整合，形成了“人工神经网络一偏最小二乘算法”，提高了算法对加和性不良的非线性体系的适应性。用这些算法辅助分光光度法，在一些吸收光谱严重重叠的多组分同时测定中(如共存稀土元素离子的同时测定)取得了令人满意的结果。这些方法赋予了专家系统强大的多元定量计算功能，在一定程度上弥补了分光光度法选择性差的弱点。 系统的上述各项功能均通过精心设计的引导界面与用户进行交流，尽可能给非专家用户以似曾相识的感觉。系统在资料的丰富程度、易用性、开放性和扩展性等方面比较令人满意，但在系统的智能化程度方面还不足。尚有一定数量的资料未及录入其中。要成为一个完备的分析专家系统，还有很多后续工作值得努力。更多还原

【Abstract】 Spectrophotometry has been widely applied to various fields for its advantages, such as sensitivity, accuracy, fast velocity, simple operation and low price of instrument. However, the color-developing reactions are of a great number of influencing factors, strong uncharacteristic feature, description model and experience model for the main, and increasingly increasing literatures for new color-developing agents and new color-developing systems, so that common users and even experts usually find that it is very troublesome to select a proper color-developing reaction and method. The normal expert system is not seemingly suitable to solving the problem in this area. Therefore, at present spectrophotometric expert system is still a blank though its expected influence and significance is tremendous. Apparently, it is a tough task with big workload and many difficulties. The thinking of research work in this paper are proposed as follows on the basis of comprehensive consideration of the characteristics of expert system and spectrophotometry:(1) Based on understanding of spectrophotometry, preceding work experience and analysis for requirement of users, summarize the fundamental rules and the key effect factors of spectrophotometry, extend its function and improve its research level.(2) Based on experimental measurement and literature search, collect the "knowledge" of expert system for various kinds of spectrophotometric systems and construct a data-base suited to express "knowledge" of expert system.(3) Put forward the quantized sieving rules of spectrophotometry.(4) To aim against the biggest weakness of poor selectivity and based on the literatures and previous research work, investigate detailedly and further improve the most practical several chemometrics methods (algorithms), and entrust a multivariate quantitative analysis function to the spectrophotometry assisted by the chemometrics methods.(5) Set up a set of practical, high efficient and open expert system suited to both common users and experts.The main object for using spectrophotometric expert system of metal elements is positioned as common users who quantitatively determine concentration or content of the metal elements in various kinds of samples by using normal spectrophotometric method. Therefore, it is necessary to refine and upgrade the experiential rules to a level that can be express with quantized mathematic language. The functions of expertsystem in the ways of selection of color-developing agents and multivariate quantitative calculation were intensified in this paper. The fundamental work in the way of sample processing was also done. In this system, the database was founded by popular office software, Access 2000, and manipulated by SQL 7.0 with Visual Basic 6.0 as main software developing tool. Many works in the way of humanized design of user guide interface were done, and multifunction, overall way, facility to use and opening way of expert system were protruded, so that the expert system could be directly used by common users without any special training, even some maintenance work such as modification and addition could be also accomplished. The process of software development was extremely difficult.The information of 202 chromogenic reagents and 1202 chromogenic systems were collected, processed and turned into database. Through a comprehensive comparison and study on the effect factors of color-developing reaction, molar absorptivity ε, contrast Δλ and acidity of color-developing reaction were selected as weighting factors in consideration of sensitivity and selectivity. The weighting values of various factors were determined and confirmed for the selected representative color-developing system by comparing the systems selected with expert experience. It was found that the results recommend by expert system were in agreement with expert experience, indicating basically reach an expert level. At the same time, some information, such as the related detailed material and absorption curves were gi更多还原