基于人工蜂群波长优选和残差增广的近红外光谱定量模型研究
Near-infrared spectral quantitative modelling using artificial bee colony algorithm based wavelength optimization and residual augmentation

提出一种基于人工蜂群(artificial bee colony，ABC)算法的光谱波长优选和残差数据增广回归算法的近红外光谱定量模型. 以勾兑果汁中苹果汁原汁含量的近红外光谱数据为基础，对原始光谱数据进行预处理，通过人工蜂群算法优选光谱波长变量，采用优选出的波长变量建立近红外光谱浓度残差增广的最小二乘回归(concentration residual augmented classical least squares，CRACLS)模型. 将ABC波长优选后建立的CRACLS模型与全光谱建立的CRACLS模型，遗传算法（genetic algorithm，GA）波长优选后建立的CRACLS，ABC波长优选后建立的PLS模型，全光谱建立的PLS模型，GA波长优选后建立的PLS模型进行比较. 实验结果表明，ABC-CRACLS模型的校正集的Rc值为0.9998，RMSEC值为0.0009，预测集的Rp值为0.9991，RMSEP值为0.0121，均优于其它几个模型. 因此提出的人工蜂群算法能够有效地处理好波长变量的优选问题，并且CRACLS模型取得良好的预测结果.
We propose a near-infrared spectral quantitative modelling method by using artificial bee colony (ABC) algorithm based wavelength optimization and residual augmentation. Firstly，near infrared spectral data of the apple juice content in blending apple juice is collected. On this basis，the spectrum signal is pretreated. Then，the ABC algorithm was used to optimize the wavelength combinations. Secondly，the concentration residual augmented classical least squares(CRACLS) model is established using the optimum combined wavelength variables for predicting the apple juice content in blending apple juice. Finally，the performances of the proposed CRACLS model with ABC wavelength optimization are compared with that of other models，including the CRACLS model with GA wavelength optimization，the CRACLS model with full spectrum，the partial least squares (PLS) model with ABC wavelength optimization，the PLS model with GA wavelength optimization，and the PLS model with full spectrum. Experimental results show that the Rc value of the calibration set of the ABC-CRACLS model is 0.9998；the RMSEC value is 0.0009；the Rp value of the prediction set is 0.9991；and the RMSEP value is 0.0121. Thus，the proposed modelling method is better than other modelling methods. Moreover，ABC can effectively deal with the optimization of the wavelength variables，and the CRACLS model has good prediction results