智能优化算法应用于近红外光谱波长选择的比较研究

发布时间：2018-06-10 00:39

本文选题：近红外光谱 + 智能优化算法　；参考：《光谱学与光谱分析》2017年01期

【摘要】：近红外光谱(NIRS)是一种间接分析技术,其应用需建立相应的校正模型。为了提高模型的解释能力、预测准确度和建模效率,需要对NIRS进行波长选择,优选最小化冗余信息。智能优化算法是以生物的行为方式或物质的运动形态为背景,经过数学抽象建立算法模型,通过迭代计算来求解组合最优化问题,其核心策略是以某种目标函数为标准,基于多元校正建模并以逐步逼近的方法筛选出有效的波长点。选用蚁群优化(ACO)、遗传优化(GA)、粒子群优化(PSO)、随机青蛙(RF)和模拟退火(SA)5种智能优化算法对烟叶总氮和烟碱近红外光谱数据进行特征波长选择,结合偏最小二乘(PLS)算法,构建了多个烟叶总氮和烟碱的校正模型,结果显示:所选用两个数据集的总氮最优模型分别为PSO-PLS和GA-PLS模型,烟碱最优模型分别为GA-PLS和SA-PLS模型,五种智能优化算法所建模型预测性能并非全部优于全谱PLS模型,但是通过智能优化算法进行波长选择后建立的PLS模型大大简化,模型的预测精度、可解释性和稳定性均有所提高。同时也对优选波长进行了解释和分析,烟叶总氮特征波长优选组合为4 587~4 878和6 700~7 200cm-1;烟叶烟碱特征波长优选组合为4 500~4 700和5 800~6 000cm-1,优选出来的特征波长具有实际物理意义。
[Abstract]:Near infrared spectroscopy (NIR) is an indirect analysis technique, and its application needs to establish the corresponding correction model. In order to improve the interpretation ability, prediction accuracy and modeling efficiency of the model, it is necessary to select the wavelength of NIRS and minimize redundant information. The intelligent optimization algorithm is based on the biological behavior or the movement of matter, establishes the algorithm model by mathematical abstraction, and solves the combinatorial optimization problem by iterative calculation. The core strategy of the intelligent optimization algorithm is to take a certain objective function as the standard. The effective wavelength points are screened by stepwise approximation based on multivariate correction modeling. Ant colony optimization (ACOO), genetic optimization (GA), particle swarm optimization (PSO), random frog (RFF) and simulated annealing (SA) were used to select the characteristic wavelengths of total nitrogen and nicotine near infrared spectral data of tobacco leaves, combined with partial least squares (PLS) algorithm. Several calibration models of total nitrogen and nicotine in tobacco leaves were constructed. The results showed that the optimal models of total nitrogen of the two data sets were PSO-PLS and GA-PLS, respectively, and the optimal models of nicotine were GA-PLS and SA-PLS, respectively. The prediction performance of the five intelligent optimization algorithms is not all better than that of the full-spectrum PLS model, but the PLS model established by the intelligent optimization algorithm is simplified greatly and the prediction accuracy of the model is greatly simplified. Interpretability and stability are improved. At the same time, the optimum wavelength of tobacco leaves was explained and analyzed. The optimum combination of the characteristic wavelength of total nitrogen of tobacco leaves was 4 587N 4 878 and 6 700 C 7 200 cm -1, and the optimum combination of characteristic wavelength of tobacco nicotine was 4 500 C 4 700 and 5 800 C 6 000 cm -1. The optimized characteristic wavelength was of practical physical significance.
【作者单位】：湖南农业大学生物科学技术学院;中南大学化学化工学院;
【基金】：国家自然科学基金项目(21275164) 湖南省研究生科研创新项目(CX2015B237)资助
【分类号】：O657.33

【相似文献】