中文摘要：

偏最小二乘回归（PLS）自带的铲椭圆图辅助分析方法具有一定的“噪音”识别能力，但无法分析多维空间中的“噪音”。在此基础上，提出将SBM算法引入到偏最小二乘辅助分析中，优化偏最小二乘回归建模。对样本数据进行综合评价，将有效的数据用来进行偏最小二乘回归，以避免“噪音”数据对回归精度的影响，弥补偏最小二乘回归辅助分析技术的不足。以中药实验实例进行计算，对于其2个因变量，SBM算法优化的PLS回归平均相对误差分别为5．0844％和8．7485％，低于直接PLS的5．5825％和9．2810％；以刀具磨损实验数据进行计算，对于其单个因变量，优化后的PLS回归平均相对误差为2．6984％，低于直接PLS的3．3526％。模拟实验结果表明，优化后的PLS回归结果比直接PLS精度更高。

英文摘要：

The ellipse assisted analysis method is provided with the ability of noise recognition in the partial least squares regression, but it is not able to analyze the ＂noise＂ in the multidimensional space. Therefore, an improved method based on SBM algorithm was proposed to optimize the partial least squares by investigating the effective sample data assessed with SBM synthetically. The lack of the aided analysis technology of the partial least squares regression was made up to avoid the effects of disturbed data on the accuracy of the regression. The average relative error of the optimized PLSR with SBM are 5. 08440/6 and 8. 7485 ~~ which is much lower than the direct PLSR＇ s 5. 5825% and 9. 2810%. In the Chinese experimental medicine, the average relative error of the optimized one is 2. 6984% which is much lower than the direct one＇s 3. 35260%. The results of simulation indicate that the improved method is better than the direct PLSR.