将移动窗口偏最小二乘(MWPLS)应用于脐橙中重金属Cd含量的激光诱导击穿光谱(LIBS)定量分析模型中,通过改变MWPLS窗口宽度并结合标准归一化处理、一阶导数、二阶导数、中心化处理和多元散射校正等5种数据前处理方法,优选与脐橙中Cd元素相关性高的光谱区间,并与传统偏最小二乘法进行对比分析。模型评价及验证结果显示,当优选移动窗口为61个波长宽度、优选区域为218.61~222.55nm时,结合一阶导数数据前处理方法所构建的模型效果最佳,验证集决定系数、预测均方根误差、主因子数、平均预测相对误差分别为0.9953,15.10×10~(-6),12,7.43%。MWPLS结合合适的数据前处理方法可以筛选出脐橙中Cd元素的LIBS光谱区域,提高定量分析模型的预测能力。 The MWPLS was applied to the laser-induced breakdown spectroscopy (LIBS) quantitative analysis model of heavy metal Cd content in navel orange. By changing the window width of MWPLS and standard normalization, the first derivative, the second order Derivative, centralization and multivariate scatter correction. Five spectral data pretreatment methods were optimized, which were highly correlated with Cd in navel orange, and compared with the traditional partial least squares method. The results of model evaluation and verification show that when the optimal moving window is 61 wavelength widths and the optimal region is 218.61 ~ 222.55nm, the model constructed by combining the first-order derivative data preprocessing method has the best effect, the validation set determination coefficient, the prediction mean square The relative error of root error, main factor and average prediction were 0.9953, 15.10 × 10 -6, 12,7.43% respectively. MWPLS combined with appropriate data pretreatment method can screen the LIBS spectral region of Cd element in navel orange and improve the predictive ability of quantitative analysis model.