本文对四种多变量校准方法———经典最小二乘法 (CLS) ,偏最小二乘法 (PLS) ,卡尔曼滤波法(KFM)以及人工神经网络法 (ANN)———在多组分浓度分析方面的性能进行了比较。选择五种红外谱图严重混叠的大气有机毒物——— 1,3 丁二烯 ,苯 ,邻二甲苯 ,氯苯和丙烯醛———作为分析对象。分别计算各种方法对该 5组分体系的平均预测误差MPE和平均相对误差MRE进行比较。结果表明 ,偏最小二乘法在处理这类问题中是最稳健的方法 In this paper, four kinds of multivariable calibration methods --- classical least squares (PLS), partial least squares (PLS), Kalman filter (KFM) and artificial neural network (ANN) Analysis of the performance were compared. Five kinds of atmospheric organic toxicants, 1,3-butadiene, benzene, o-xylene, chlorobenzene and acrolein, which are seriously alias in the infrared spectra were selected as the analysis objects. Various methods were calculated to compare the average prediction error MPE and the average relative error MRE for the five-component system. The results show that PLS is the most robust method in dealing with such problems