波长选择算法在近红外光谱法中药有效成分测量中的应用

建立基于仪器分析方法的质量标准是中药进入国际市场的必要条件。近红外光谱技术以其能够反映样品的多种信息、易于在线应用的优势，应用于中成药生产的在线质量监控，可以提高中成药的质量控制标准，加快中药现代化的进程。但在近红外光谱检测中存在着各成分谱图重叠严重，光谱信息冗余，特征吸收区域不明显的问题，需要对采集到的波长进行优选，以达到提高模型预测精度和简化模型的目的。从近红外光谱方法测量中药有效成分的基础研究入手，以冰片含量的检测为例，尝试采用遗传算法与模拟退火算法结合的模拟退火遗传算法及物理意义相对明确的多链逐步选择法对校正模型的波长进行优选。结果表明，波长选择的方法可以使模型采用的波长数减少的同时提高预测精度，波长选择最多可将波长数减少84％，预测精度提高47.6％。

Application of Wavelength Selection Algorithm to Measure the Effective Component of Chinese Medicine Based on Near-Infrared Spectroscopy

Near infrared (NIR) spectroscopy has raised a lot of interest in the pharmaceutical industry because it is a rapid and cost-effective analytical type of spectroscopy with no need for extensive sample preparation, and with the easy-realizable ability of on-line application. The NIR technology can increase the quality control standard of the Chinese medicine and accelerate the entry into the international market. In the present paper, two methods for wavelength selection are applied to the measurement of borneol, one of which is the multiple-chain stepwise, which tends to select many variables in the same area containing valuable information, and the other is the mixture genetic algorithm, which incorporates simulated annealing so as to improve the local searching ability while maintaining the global searching ability. The results present that the number of wavelength is reduced to 16% compared with the original number of wavelength, and the prediction accuracy has increased 47.6%. Therefore, the method of wavelength selection is a good way to enhance the prediction accuracy and simplify the model in NIR region.