NTKL法可以同时检测光谱异常点、参考数据异常点及其他们共同的异常点，该方法的基本思路是基于蒙特卡洛采样，随机将样本集划分为校正集与预测集，用校正集建模，并对没参与建模的样本进行预测误差计算，进行几千次这样随机采样，每个样本点都可得到其预测误差分布。如果奇异样本在校正集中，整个模型的质量将受到影响；相反，如果奇异样本在预测集中，仅此样本的预测结果受到影响。-NTKL method can simultaneously detect abnormal spectrum and abnormal reference data points and their common abnormal point, the basic idea of this method is based on the monte carlo sampling, sample set random be divided into calibration set and prediction set, using correction modeling, modeling and participated in the samples of the prediction error calculation, a thousands of times that of random sampling, each sample point can be obtained the prediction error distribution.If singular samples in correction, the model of quality will be affected.Singular samples in predicting concentration, on the contrary, if only for the sample prediction results are affected.