【作者】 王红霞；

【导师】 王洪艳；

【作者基本信息】 吉林大学 ， 分析化学， 2005， 硕士

【摘要】 本文在传统的BP 网络的基础上增添了一个拟合网络,创新地提出了双网络ANN 算法。用拟合网络拟合出的吸光度值与实测吸光度值间的误差作为学习样本选择的依据,解决了学习样本选择的难题,大大提高了人工神经网络预测的准确度。在中草药样品有效成分的测定中,首次提出了虚拟组分排除法,该法与双网络ANN 结合,可以不经化学分离去除干扰组分,实现了中草药秦皮中有效成分秦皮甲素、乙素的同时测定,取得了良好的效果。对22 个秦皮样品进行了验证,结果表明,双网络ANN 对于秦皮甲素、乙素预测的准确度可以达到91.0%以上。另外,将卡尔曼滤波算法与流动注射-化学发光法结合实现了复杂地质样品中化学性质极为相近的铌、钽元素的同时测定。该方法分析快速,仪器简单,操作简便,是野外现场分析很好的研究手段。方法的检出限为Nb（V） 2.1×10^-3 μg g^-1,Ta（V） 4.0×10^-3 μg g^-1,精密度为Nb（V）4.9%,Ta（V）3.3% （n = 9）。更多还原

【Abstract】 Chemometrics is an interdisciplinary subject. It includes statistics,mathematics and mathematical statistics. In recent years, with thedevelopment of computer science, more and more people attach importanceto chemometrics. Multi-components simultaneous determinations in thecomplicated analytical system, without chemical separation, are the mainresearch objects of analytical chemists, which is also a fast developingresearch field in chemometrics. In this dissertation, double artificial neuralnetwork analysis-UV visible spectroscopy method was used to simultaneousdetermination of aesculin and aesculetin in Chinese herbal medicine-CortexFraxini, and Kalman Filter analysis flow-injection chemiluminescence(FI-CL) method was developed to simultaneously determine trace amountsof niobium and tantalum in geological samples. The main research work isdescribed as follows.1. Simultaneous determinations of aesculin and aesculetin in Chineseherbal medicine-Cortex Fraxini by double ANN and Uv-vis.A Chinese herbal medicine always concludes many chemical components,which is a typically complicated system. The analytical methods are usuallyHigh Performance Liquid Chromatography (HPLC) and Thin LayerChromatography (TLC).The former method suffers from high-costinstrument, complex operation, and long-time analysis, and the latter one ispoor recurrence. HPLC can provide an effective method to determinecomponents in Chinese herbal medicine. However, it will not take effectwithout compared samples which is expensive. In this article, we attemptedto analyze the UV-vis data of Chinese herbal medicine by the improveddouble ANN, and realized the simultaneous determination of maincomponents by self-learning, self-revision and self-fit, without the boringprocess of separation.The revised double ANN, based on the traditional artificial neural network,adds the function of self-distinguishing and self-fit, and provides the basisof selecting the learning samples. The calculating process includes twonetworks. (1) Learning network. In this section, ANN predicts components’content, using the net parameters gotten by any learning sample. (2) Fitnetwork. Take the components’content as the input value of the fit network,and regard the absorbency value as the standard value to start the learningprocess until the relative error between the experimental value andpredictive value(Uv) is smaller than the admitted number. It has beenproved that the combined method of double ANN and UV-vis cansimultaneously determine aesculin and aesculetin in Cortex Fraxini. Thedetermination accuracy was above 91.0% when the admitted relative errorwas 15%.Because of the complexity, the Chinese herbal medicine always includesmany components. Therefore, other interference components will influencethe determination. To solve this problem, we first put forward the virtualcomponents elimination method. The purpose is adding an unknown valueto the learning samples, and making it to involve the learning and fit process.If the Uv value is larger than the admitted error (E), adjust the value ofvirtual component, and continue to learn and calculate in double ANN.when the Uv value is smaller than E, ANN will forecast the results andreport them.In addition, the precision of double ANN was tested. The determinationprecision of aesculin and aesculetin was 1.0% and 0.7%, respectively. Thecombination of double ANN and the virtual components elimination method,without the separation, can successfully determine effective components inChinese herbal medicine.2. Fast simultaneous determination of niobium and tantalum byKalman Filter analysis with flow-injection chemiluminescence method.Niobium (Nb) and tantalum (Ta) are widely known to be extreme similar inchemical characteristics, which results in a difficulty to separate anddetermine them. Although inductively coupled plasma mass spectrometry(ICP-MS) is convenient for the simultaneous determination of Nb(V) andTa(V), it is not suitable for field scene analysis. Therefore, there is a strongrequirement for a simple, fast and effective method to determine trace Nb(V)and Ta(V) in order to meet the need of field scene fast analysis.Flow-injection chemiluminescence (FI-CL) is an available trace-analysistechnique. However, the fatal disadvantage of the chemiluminescencemethod is its poor selectivity, which makes it difficult to differentiate thechemiluminescence signal of Nb(V) and Ta(V). The general method is更多还原