基于遗传算法的近红外光谱建模样品集优化研究

Study on Optimization of NIR Spectroscopy Calibration Set for Soluble Solid Content in Nanfeng Mandarin Fruit Based on GA

探讨遗传算法优化方法在南丰蜜橘内部品质可溶性固形物的建模集样品选择中的应用,根据优化结果建立了南丰蜜橘可溶性固形物的快速无损检测数学模型。研究选择的波长范围为350～1 800nm,应用遗传算法优化建模集样品,优化后建模集样品数量由89个减少到36个,累积变异系数达到99.99%,交互验证均方根误差达到最小。研究结果表明:优化后的建模样品数量大大减少,且保证了南丰蜜橘可溶性固形物的快速无损检测数学模型的稳定性。

The objective of this research is to optimize the sample of calibration set by genetic algorithm （GA） for determining soluble solids contend （SSC） of Nanfeng mandarin fruit, then mathematical models are built for rapid determining SSC of Nanfeng mandarin fruit based on the optimization. The samples of calibration set are optimized by GA in the wavelength range of 350 - 1 800 nm. After optimizations, the sample number of calibration set decreases from 89 to 36, accumulative variance value reaches 99.99%, and the least root mean square error of cross validation （RMSECV） is obtained. The research results show the sample number of calibration set decreases vastly, and the stability of model for rapid determining SSC nondestructively is ensured in the Nanfeng mandarin fruit.