摘要
对模拟退火算法进行了优化,并以小麦粉样品的光谱为对象,利用优化后的算法在分子含氢基团一倍频区域1400—1860nm(7144—5376cm-1)优选出了分析小麦蛋白质含量的4个特征波长,结合多元线性回归(MLR)建立了定标模型。同时,利用这个谱区的全谱数据,结合偏最小二乘法(PLS)建立了另外一个模型。经过比较,发现这两个模型具有相同的预测效果。说明优化后的模拟退火算法能很好地用于近红外光谱分析中定标波长的选取,进而建立“精而简”的模型。这对于简化定标模型及确定分立波长型仪器的定标波长具有十分重要的价值。同时,对处理其他优化组合问题也有一定的指导意义。
The simulated annealing algorithm （SAA） was optimized. Four characteristic wavelengths were selected in the first overtone（7144-5376cm^-1）of bonds involving hydrogen using the optimized SAA to the protein of wheat powder. A calibration model was obtained by MLR with these four wavelengths. The other calibration model was established by PLS in this region directly. The prediction results carried out the same quality to the two models. The optimized SAA was powerful to simplify the calculation of calibration and to locate the wavelengths of calibration for discrete scanning spectrometers, it is also useful for determining other optimal combinations.
出处
《光谱实验室》
CAS
CSCD
2006年第5期921-925,共5页
Chinese Journal of Spectroscopy Laboratory
关键词
模拟退火
偏最小二乘
近红外光谱
波长选择
Simulated Annealing, Partial Least Squared, Near Infrared Spectrum, Wavelengths Selection.
