小波特征提取和随机森林模型解析色谱重叠峰  

Resolution of Overlapped Chromatographic Peaks by Wavelet Feature Extraction and Random Forest Model

作　　者：张鹏程[1] 王爱民[1] ZHANG Peng-cheng;WANG Ai-min(School of Instrument Science and Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]东南大学仪器科学与工程学院,江苏南京210096

出　　处：《测控技术》2019年第5期31-35,共5页Measurement & Control Technology

摘　　要：针对神经网络算法在当前色谱重叠峰解析领域存在易过拟合、网络结构复杂、学习效率低等问题,引入了随机森林模型。利用gausl小波模拟原始信号导数,选取合适的尺度并提取信号的特征拐点;以特征点作为模型输入、子峰面积比作为输出,使用随机森林模型拟合两者之间的映射关系;采用交叉验证的方式确定随机森林模型的参数,并使用CART算法进行模型的构建和训练;一系列实验与现有方法的对比,证明了本文方法不仅能准确对特征拐点和子峰面积之间进行拟合,在模型训练时间上还具有很高的效率。The random forest model was introduced to overcome the shortcomings of neural network in overlapped chromatographic peak resolution, such as over-fitting, complex network structure and low learning efficiency. Gausl wavelet was used to simulate the derivative and calculate the inflection points of the chromatogram by selecting the right scale. The feature points and sub-peak area ratios were served as input and output of the random forest model, which used cross-validation to determine the initialization parameters and the CART algorithm to construct and train the model. The comparison between a series of experiments and the existing methods proves that the proposed method can not only accurately fit the inflection point and the sub-peak area, but also improve the efficiency in model training.

关 键 词：重叠峰解析 小波变换 随机森林模型 交叉验证 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置]