利用基于小波特征提取的网络模型解析色谱重叠峰  

The application of network model based on wavelet feature extraction in resolution of overlapped chromatographic peaks

作　　者：王佩佩[1] 宋晓峰[1] 杨平[1] 

机构地区：[1]南京航空航天大学生物医学工程系,江苏南京210016

出　　处：《计算机与应用化学》2007年第5期673-677,共5页Computers and Applied Chemistry

摘　　要：提出了一种新的色谱重叠峰解析模型——基于小波特征提取的RBF神经网络模型。首先利用小波变换奇异性的检测原理,从原始色谱信号中提取特征点,这些特征点蕴含了反映色谱峰形状的信息,包括重叠峰个数、保留时间等信息。由小波变换获得的特征点来确定RBF网络的隐节点数目和网络参数的初值,即将拐点对数作为隐节点数目,将峰宽估计值作为输出层连接权的初值,将峰高估计值作为隐节点宽度的初值。再用RBF网络来拟合原始重叠色谱信号,梯度下降法训练后获得的网络参数作为解析结果,实现了重叠色谱峰的分离。实验结果表明:本方法快速、准确、可靠,能有效解析未知组分数的重叠峰。A new model-resolution of overlapping chromatographic peaks of unknown components number is presented in this paper. The singularity detection principle of wavelet transform was used to extract characteristic points which included the information of chromatographic peak shape, such as the number of overlapped peaks and retention time. According to extractive characteristic points, the number of hidden layer nodes and initial values of the parameters were estimated. Inflexion number decided hidden nodes number, peak width decided initial value of connection weight, peak height decided initial value of hidden node width. Radial basis function （RBF） neural network was employed to fit overlapped chromatographic signals. The analytica results were form network parameters after training network with gradient descent method. Experimental results indicate that the proposed method has good performance with fast training speed, high accuracy and reliability, and can resolve overlapped peaks which has unexpected number of components.

关 键 词：小波变换 径向基神经网络 重叠峰解析 色谱 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] O65[自动化与计算机技术—控制科学与工程]