基于遗传算法的近红外光谱谱区选择方法

Region Selection Method of Near Infrared Spectrum Based on Genetic Algorithm

针对以每个波长一个染色体基因的 WSGA方法在波长数目较大时搜索空间太大的问题 ,提出了基于遗传算法的近红外光谱谱区选择方法 (RSGA ) ,把全部谱区平均分为多个子区间 ,各子区间的所有不同组合构成搜索空间 ,使用 R/(1+δ)最大作为优化目标 ,使用遗传算法寻找一个最佳子区间组合 ,作为最后参与建模的光谱谱区。计算实例表明 :经 RSGA优化选择谱区后 ,不仅波长点数减少 ,而且 PL S L OO CV预测值与标准值的相关系数 R得以提高 ,交叉检验预测均方差得以减少 ,从而可以减小建模运算时间 ,剔除噪声过大的谱区 ,使最终建立的农产品品质检测近红外光谱模型的预测能力和精度更高。

An efficient method named region selecting by genetic algorithms (RSGA) for building a PCR or PLS calibration model of NIR was presented in this paper, in which each gene of chromosome no longer represented a wavelength point only like the approach of wavelength selecting by genetic algorithms (WSGA) did. In the RSGA method, one needs to divide averagely the full spectral band into many sub-regions, and to build a research space with all the combinations of these sub-regions, but not a single wavelength point. And then regarding maximizing R (the correlation coefficient of prediction values and standard values) and minimizing the root mean squared error of prediction of cross validation (RMSPCV) of residuals as the optimal object, regarding maximizing R/(1+δ) as the object function, it optimizes the research space by the genetic algorithm. Finally, a PCR or PLS calibration model of NIR is built by using the optimal combinations of these sub-regions. The calculated instance shows that, using RSGA method, the wavelength points and the run time of model building are reduced, R is increased, and RMSPCV is decreased, which lead to a more accurate model.