Abstract: Wavelength selection of spectral analysis for substances content in the liquid is studied. A multi model fusion algorithm is proposed to improve the prediction accuracy based on reducing the number of wavelengths. An adaptive weighted wavelength selection algorithm（iCARS） is proposed in this paper. Multi model fusion with iPLS and CARS combined with variable selection and vertical and adaptive optimization sample selection in the wavelength selection. And through wavelength band selection and wavelength selection by step the algorithm achieve wavelength optimization selection. Through the MATLAB simulation by using the PLS model for forecasting models in the original beer wort concentration prediction and experimental, iCARS algorithm selects 19 variables. By using these variables PLS prediction model get RMSEP 0.139. This algorithm makes the number of the variables significantly reduced, and the prediction ability is improved obviously. It can implement the optimal selection of variables.

Key words: multi-model?fusion, wavelength selection, partial least squares, adaptive

摘要： 针对液体中物质含量的光谱分析中波长选择展开研究，构建一种多模型融合算法进行波长选择，在减少波长数量的基础上提高预测精度。提出的区间自适应加权波长选择算法（iCARS），应用iPLS算法和CARS算法的多模型融合，在波长选择中结合变量与样本纵横向自适应优化选择，以及对波段和波长分步骤选择实现波长优化选择，实现了波段和波长的自适应选择。通过MATLAB仿真实验，将PLS模型做为预测模型，在对啤酒原麦汁浓度预测仿真实验中，应用iCARS算法选择19个变量，利用这些变量建立PLS预测模型得到的RMSEP为0.139。变量数目大大减少，预测能力明显提高，实现了变量的优化选择。

关键词: 多模型融合, 波长选择, 偏最小二乘, 自适应