| Since the 20th century, due to the development of modern science and technology, the development of analytical chemistry experienced three great changes, so that modern analytical chemistry has developed into a subject of which instrumental analysis is the main stream. Drug analysis is an important component of analytical chemistry as well as of the drug science. New demands are proposed as the result of the development of drug science and the human pursuit of quality of drugs. In recent years, due to the rapid development of science and technology and mutual infiltration of subjects, drug analysis has developed into instrument-analysis-based modern drug analysis. The main task of Modern drug analysis is to effectively control the quality of drug raw materials and preparations in the processes of drug development, production, distribution and use by means of accurate and rapid analysis to ensure the safety and effectiveness of drug use. The advent of the Information Age, of which computer application is the main indicator, accelerates the development of analysis technology, especially the development of instrument analysis become more rapidly, which not only provides a variety of very sensitive, accurate and rapid means for the drug analysis subject, but also by getting into the body, monitors dynamically in the process of reaction. Drug analysis develops towards a more precise, fast, convenient, trace, and environmental protection way.With the advent of the 21st century, the medical model has gone through the transition from treatment to prevention. At present, the development of pharmacy at home and abroad is very rapid, firstly, new drugs are launched, and widely used in clinical, and secondly, antibiotic medicines replacement cycle was shortened, and thirdly, the old drugs were used in new different ways. The main reason is that the basic pharmaceutical research was draw wide attention attached great importance by governments as well as the result of the close cooperation of doctors and pharmacy workers studies. Therefore, the urgent task the drug workers are facing is no longer just static conventional drug testing, but to go into the process of drug production, and to achieve real-time, in situ, and online quality control to drug raw materials, intermediates, semi-finished and finished products.Nowadays in the quantitative analysis of drugs, it is commonly used chromatography (GC and HPLC) and spectroscopy (IR and UV) as the pharmacopoeia standard method and the Ministry of Health universal standard method. However, these methods, without exception, must dissolve samples, isolate its ingredients, and then further measure the content of the ingredients, which is time-consuming and laborious. The thesis will focus on using SVM-NIR spectrometry, for the analysis of drug quality, that is, non-destructive rapid monitoring the quality of medicines in order to achieve real-time online monitoring of drug quality.NIR spectroscopy method is a fast and efficient online testing technology, which is used to do multi-component analysis at the same time by the multivariate calibration model. Since the 1990s NIR has developed rapidly and been the most attractive spectrometry technology, which is the integration of the spectral measurement technology and stoichiometry, known as the Giant of the analysis. One of its key steps is the establishment of multivariate calibration model on NIR spectroscopy measurement data. Typically, only when we get a large number of known samples that is well-distributed, can we gain the spectrum analysis accurately. In the practical analytical work, however, medicines and other objects are often difficult to obtain a large number of known samples, and there always exists nonlinear mapping between the spectral data and concentration of the component. Thus the study of nonlinear modeling of fewer samples is the direction and the focus of the NIR spectroscopy analysis. Recently, many researchers apply artificial neural network (ANN) method to establish the nonlinear regression model of NIR spectroscopy, under the condition of more known samples, we can get a model that can correct and anticipate the relatively smaller errors by means of ANN. However, in the field of drug analysis most of the data modeling issue is a mathematical small sample of ill-posed problem, and traditional methods ,such as linear and non-linear regression, artificial neural networks ignored this characteristic, often view them as endless samples , well posed problem , data modeling has led to the "overfitting" and poor generalization, that is the problem of better fitting for known samples but worse for the unknown samples, but may cause obstacles in practice.The support vector machines (SVM) founded by Vapnik, performed a unique characteristics and superiority on the promotion in the machine learning, has been extensive studied and applied. And different from the conventional neural network based on the experience of risk minimization principle (ERM), SVM is based on statistical learning theory of structural risk minimization principle, which is strictly based on the theory and mathematics. The core idea of SVM is that the input space through specific function mapped to the high-dimensional feature space, constructed classification hyperplane that can maximize the interval category and decision-making rules so as to control model capacity (complexity), inhibit overfitting, and improve the performance of promoting model, by SVM it we can solve the non-linear, high-dimension, the local minimum point and other problems, and small sample studies make it a strong generalization.SVM has four notable features:1). with the idea of large interval to reduce the VC-dimensional (Vapnik-Chervonenkis Dimension) of classifier, to implement structural risk minimization principle, to control capacity of the promotion of classification, and to maximize the generalization of the learning ability;2). With Mercer kernel to achieve non-linear and linear algorithm. A major feature of SVM is that nonlinear problems are mapped to linear problems of the high-dimensional space, which is replaced by a kernel function in the plot operation, thus avoiding the high-dimensional operations, ingeniously solved the nonlinear problems;3). It is sparse, that is the coefficient of small sample (SV) can be any but zero. In the aspect of the promotion, the less support vector is, the better the promotion capacity is in terms of statistical significance. And in terms of calculation, support vector reduces the calculation of the kernel form discriminant;4). Algorithm concerns of convex optimization, thus local optimal solution must be global optimal solution and avoids the result of multi-solutions.The above features are not shared by other learning algorithm such as neural network learning algorithm.In this thesis, considering the complex characteristics of pharmaceutical composition and the superiority of establishing non-linear model by means of support vector regression, we combined SVM and NIR in the quantitative analysis of its active ingredients, forecasted active ingredients of three different drugs in a quantitative way, thus established a practical forecasting model, and achieved the real non-destructive quantitative analysis. This paper applies root mean square error of cross-validation (RMSECV) to estimate cross-validation, and relative standard error (RSE) to measure the performance and the predictive abilities of the model, and it also calculates the cross-validation and the correlation coefficient (R) of the Test set.Compared with the conventional drug test method, this method can make a quantitative analysis of drug quickly and accurately, particularly applicable to the online quality control of drugs; meanwhile in the access of quantitative analysis it did not undermine the composition of medicines, undoubtedly this method means a lot for some expensive drugs . In view of the large numbers of subjects in the statistics field of small samples of drugs, SVM in drug analysis is very promising; therefore the application of SVM theory to solve the practical problems is of great significance.
|
PDF Full Text Request