CASIA OpenIR  > 毕业生  > 博士学位论文
红外光谱定量分析建模算法研究
其他题名Novel Multivariate Calibration Algorithms for Infrared Spectroscopic Data Analysis
毕一鸣
学位类型工学博士
导师彭思龙
2014-05-21
学位授予单位中国科学院大学
学位授予地点中国科学院自动化研究所
学位专业模式识别与智能系统
关键词偏最小二乘 化学计量学 红外光谱 多元校正分析 定量分析 Partial Least Squares Chemometrics Infrared Spectra Multivariate Calibration Quantitative Analysis
摘要摘要 红外光谱技术在过去二十年来得到了长足的发展。红外光谱技术可以对生产过程进行分析和监控,因此被广泛的应用到制药,食品,石油化工等诸多领域。红外光谱技术(包括近红外和中红外)的主要优点是包含信息丰富,采集快速,无损且采样成本低。多元校正技术是红外光谱定量分析的主要技术之一。然而,如何在建模中提取有效信息,提出光谱中的噪声和干扰,依然是红外光谱定量分析中亟待解决的问题。在本论文中,围绕提高模型的精度,鲁棒性和解释能力,提出几种红外光谱定量分析的新算法,包括: (1)提出一种散射校正算法用于近红外光谱定量分析。该算法通过定义并寻找光谱中的干扰主导区域,利用这一区域估计该样本的散射校正参数并应用到整条光谱中进行校正,从而避免了光谱吸收区域对校正参数估计的影响。算法分为两个步骤:第一步是通过筛选若干光谱区域,找出最优的干扰主导区域;第二步是通过选出的干扰主导区域来估计校正参数,并用这些参数在全谱上进行散射校正。在校正完成后,利用校正后的数据进行PLS建模和预测。相比传统算法,所提出的算法利用了待测物浓度信息来帮助寻找干扰主导区域。该算法的另一个优点是可以成功地应用于多组分数据,寻找各组分间共同的干扰主导区域,从而实现各组分在建模前实现统一的散射校正。在公开数据集中的实验表明所提方法比传统方法在建模精度上有显著提高。 (2)提出一种新的权值更新方法,用以提高PLS模型的模型精度和解释能力。针对PLS算法流程中权值向量易受光谱噪声干扰的问题,提出使用一种分段线性映射函数(SLT变换)对权值向量进行更新,利用更新后的权值向量进行原算法中各载荷向量,得分向量的计算。在保持原有PLS流程不变的情况下,通过这种处理可以降低原算法中由于光谱噪声等干扰造成的投影向量估计不准确等问题。我们的实验证明,提出的SLT-PLS算法可以降低原有模型的潜变量数,在使用较少潜变量的情况下即可收敛,这对提高PLS模型的解释能力很有帮助。此外,光谱噪声影响较大的情况下,SLT-PLS可以显著提高PLS模型的预测精度。 (3)提出一种利用PLS和集成学习思想的红外光谱定量分析算法,称为双堆PLS算法(DSPLS)。算法包括对PLS模型的两层集成:通过对不同光谱区域的信息分别建模并集成,称为内层集成;对不同的内层模型进行二次集成,称为外层集成。内层集成通过权重的调整来降低光谱噪声和其他干扰区域对模型的影响。外层模型是针对PLS中不恰当参数选择导致的过拟合问题,在训练集样本不充分大的情况下,通过多模型融合来回避最优参数选择,增强模型的鲁棒性。同时,在外层集成中,我们提出一种选择性集成的方式,利用部分模型集成,得到比全部模型集成更好的建模效果。实验表明,相比传统算法,新算法有更好的模型精度和鲁棒性。
其他摘要Abstract In the past two decades, there has been considerable research interest in utilizing infrared techniques for process analysis, monitoring and for the control of broad biological and chemical fields, such as pharmaceutical, food and petroleum industries. The infrared techniques, including Near-infrared and Mid-infrared technique, is widely used in these areas, extracting information with fast, noninvasive and low cost of sample measurement. Multivariate calibration techniques such as Partial Least Squares (PLS) was often used because of the ability of overcome both the dimensionality and the collinear problems of infrared spectral data. However, there are still a number of obstacles with infrared quantitative calibration, such as how to extract useful information and eliminate noise interference concurrently during use. In this sense, the research in this thesis focuses on the quality improvement of the spectra and contributes to improving the robustness of the quantitative analysis model. The content in this thesis includes the following aspects: Scattering correction of the infrared spectra are explored in this paper. a novel algorithm was proposed to find a spectral region automatically that the interesting chemical absorbance and noise are low, that is, finding an interference dominant region (IDR). Based on the definition of IDR, a two-step method was proposed to find the optimal IDR and the corresponding correction parameters estimated from IDR. Finally, the correction was performed to the full spectral range using previously obtained parameters for the calibration set and test set, respectively. Contrary to conventional correction methods, the proposed method used the infromation of reference values to find IDR. In addition, the method can be applied to multi target systems with one IDR suitable for all targeted analytes. Tested on two benchmark data sets of near-infrared transmittance spectra, the performance of the proposed method provided considerable improvement compared with full spectral estimation methods and comparable with other stat-of-art methods. A modified partial least squares algorithm is presented based on a novel weight updating strategy. In original PLS, the estimation of weight vector are often suffered with X-space noise, led to complex model and poor performance. To slove this problem, the slice transform (SLT) technique is introduced to provide a piece-wise linear mapping to the weight vectors. The new weight can ha...
馆藏号XWLW1976
其他标识符201018014628025
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6583
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
毕一鸣. 红外光谱定量分析建模算法研究[D]. 中国科学院自动化研究所. 中国科学院大学,2014.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20101801462802(3640KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[毕一鸣]的文章
百度学术
百度学术中相似的文章
[毕一鸣]的文章
必应学术
必应学术中相似的文章
[毕一鸣]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。