K均值改进留一校验法在煤炭近红外光谱异常样本剔除中的应用研究

Application research of improved K-means leave one out method in rejecting of abnormal samples of coal near infrared spectrum

针对现有留一校验法存在剔除异常样本耗时长、误判的缺陷,提出一种K均值改进留一校验法,并将其用于煤质分析中异常样本的检测与剔除。该方法首先利用K均值聚类法对样本进行聚类,得到可疑样本;然后将可疑样本作为验证集,通过留一校验法进行二次判别,剔除异常样本。实验结果表明,K均值改进留一校验法能快速、准确剔除异常样本,提高了模型的预测精度。

In view of problems of time-consumption,misjudgment of rejecting abnormal sample existed in current leave one out method,an improved K-means leave one out method was put forward for detecting and eliminating abnormal sample in coal quality analysis.Firstly,the method uses K-means clustering method to cluster samples,and gets suspicious samples;then it takes suspicious samples as a validation set,and adopts leave one out method to do quadratic distinguishing,so as to eliminate abnormal samples.The experimental results show that the K-means leave one out method can eliminate abnormal samples quickly and accurately,and improves prediction accuracy of models.