这是一种基于近红外光谱的非线性建模方法及系统，从各所述近红外光谱数据随机挑出一部分作为校正集，挑出一部分作为验证集；将所述校正集和所述验证集通过主成分分析得到光谱特征空间；在所述光谱特征空间中，通过马氏距离法选取所述校正集里与所述验证集的各个样本最近似的样本作为校正子集；从所述校正子集中提取主成分数，作为BP神经网络的输入层建立回归模型，不仅能解决各因素之间多重相关的问题，还避免了大量的噪声和一些无用的信息，降低了变量维数，在BP神经网络的非线性映射能力和适应学习能力的基础上，提高了模型的预测稳定性和精度。-This is a kind of nonlinear modeling method and system based on near infrared spectrum, described　the near infrared spectrum data randomly selected part as calibrating, pick out the part as a validation set Will be described in the calibration set and described in the validation set is obtained by principal component analysis (spectral feature space It is spectral feature space, the selection method described by markov distance calibration set and validation set described in each sample with the sample as the calibration subsets Principal components extracted　calibration described subset, as BP neural network input layer to establish a regression model, not only can solve the problem of multiple correlation among various factors, also avoid a lot of noise and some useless information, reducing the variable dimension, the nonlinear mapping ability of BP neural network and adaptive learning ability, on the basis of improve stability and accuracy of the prediction of the model.