Applications Of Wavelet Transform And Artificial Neural Networks On Chemistry

Zhang wenjun, male, was born in 1969, and go in for the study on chemometrics from 1996, supervised by Prof. Xu Lu. The thesis is divided into two parts: Applications of wavelet transform and artificial neural networks on chemistry. Wavelet transform is a new mathematical method. Recently, it was applied on chemistry extensively. In this thesis, an introduction of wavelet transform and multi-resolution analysis is presented. Three data compression methods based on wavelet transform for spectral informations have been developed. Using the multi-resolution analysis, we compressed spectral data by Daubechies's compactly supported orthogonal and orthogonal cubic splines wavelet. By using orthogonal cubic splines wavelet and storing very few large coefficients, we have achieved a favorable data compression. We also used the methods of wavelet transform to denoise the singles of Surface Plasmon Resonance, and a good result was achieved. The applications of artificial neural networks in chemistry have been investigated extensively. There are many factors that affect the result of artificial neural networks, in which, variable selection is the most important. Including too many variables will often cause a substantial reduction in the predictive ability of the model to the predictive set. So that it is necessary to select a set of descriptors that produces the most predictive model. Several strategies for variable selection have been suggested. In this paper, a comparison of the different methods was performed. These methods include classical algorithms such as forward selection, backward elimination, and stepwise procedure; leaps and bounds regression; method of orthogonal descriptors as well as the new optimization technique, genetic algorithm. Some interesting hints were obtained. Furthermore, we suggested a new algorithm that combines the algorithm of forward selection and combinatorial algorithm. The good results can be obtained with the new algorithm; the local minimal can be avoided in some degree. In this paper, neural networks also were applied to quantitative structure-property relationships analysis: 1) Studies on the toxicity of aniline, nitrobenzene and phenol derivatives and their structures. 2) Studies on HEPT compounds for their anti-aids. 3) Studies on nitrogen mustards and 2-formylpyridine thiosemicarbazone compounds for their anti-cancer. 4) Studies on phenol and aniline derivatives for their gas chromatographic retention values. 5) Studies on discrimination of tea by Self-organizing feature map network.