【作者】 杨小曼；

【导师】 郑建斌；

【作者基本信息】 西北大学 ， 分析化学， 2003， 硕士

【摘要】 本论文研究了傅立叶自去卷积、小波变换、神经网络、遗传算法及其结合用于解决化学信号中有用信息的提取、去噪、重叠峰分辨等问题的可能性，从而扩大了小波变换、神经网络、遗传算法的应用领域，发展了化学计量学方法。全文共分四章，创新性主要有以下四方面： 1、将小波变换与傅里叶变换结合用于重叠峰分辨方法的研究，提出了频域小波变换Fourier自去卷积法。与其他FSD方法相比较，频域小波变换Fourier自去卷积法具有更好的重叠峰分辨效果。这主要是由于从傅里叶变换得到的模R与其经小波变换获得的模糊项C_N具有相似的线性和峰宽，能在较大程度的与原始谱峰相符。 2、根据示波计时电位信号的自身特点，提出了一种示波信号能谱的Fourier自去卷积法。该法使用支持电解质溶液产生的示波计时电位信号作为线性函数，去极剂本身的示波信号作为去噪函数。该方法不仅可以有效的提高检测的灵敏度、减小FSD过程中的噪音干扰、防止过去卷积或去卷积不足，还具有线性函数选择简便、不用选择去噪函数及线性函数的峰面积(A)和半峰宽(σ)等优点，从而使得去卷积操作更为简单易行，且避免了FSD过程中人为因素的引入。 3、将BP神经网络用于从强噪声背景中提取微弱信号并用于含较强噪音的示波计时电位信号的有用信息提取，研究了信号峰峰宽和噪音含量对提取结果的影响。 4、将遗传小波神经网络用于极谱信号的滤噪和压缩。由于使用遗传算法优化神经网络的参数，避免了网络陷入局部优化和网络参数选择时的人工参与，提高了神经网络处理化学信号的智能化程度。更多还原

【Abstract】 In this thesis, the possibility of Fourier self-deconvolution(FSD), wavelet transform(WT), artificial neural network(ANN), genetic algorithms(GAs) and their combination to solving problems such as feature extraction of signals, de-noising and resolution of overlapped peaks have been explored. The application fields of FSD, WT, ANN and GAs are enlarged and some new chemometrics methods are founded. The thesis consists of four chapters, and the author’s contributions are in the following four aspects:1.Combinating WT and Fourier transform, a new FSD with wavelet transform in frequency domain is presented. Compared other FSD, the new method can effectively resolve overlapped peaks. This is because the module R obtained after FFT for the original signal is similar to appormix CN obtained from WT for the module R in their linearity and peak width.2.A new method of FSD, FSD of the power spectrum of oscillographic signal, is proposed. In this method the oscillographic signal of base solution and depolarizer are used as line shape function and filter function, respectively. The new method can not only increase the relative incision depth, improve sensitivity and the resolution of the overlapped incision, avoid overdeconvolution and underdeconvolution, but also eliminatethe need for choosing line shape function and filter function. This avoid the subjective interference factors and FSD operation is made easy.3. The BP ANN is applied in extracting weak signal and information of oscillographic chronopotentiometric signals from strong noise background. Effects of noise and peak width of signal are studied.4.Genetic wavelet neural network(GWNN) is proposed and applied to thecompression and de-noising of simulated signal and polargraphic signal. The improper selection of network parameters and local optimal solution which often occurs in the training process of WNN is avoided because the parameter of WNN is optimized by GAs. The intellectualized level of artificial neural network applied in chemical signal processing has been improved ulteriorly.更多还原