Recently, the technologies for detecting abrupt changes have been prevalent in many fields. However, as the conven-tional strategy, parametric method has some inherent limitations when it is used in real production environment. In order to solve these problems, we propose a novel nonparametric method for abrupt changes detection, which solves the problem by AUC ( Area under the Curve) . We divided the process into two phases:during the pre-analysis period, we set two continuous scanning win-dows and then calculate the AUC values of the samples in difficult scanning windows in order to obtain the means and variances;during the detection period, we detect the changes by hypothesis testing. In simulation experiments, the method we proposed out-performs CUSUM in terms of robustness and multiple changes detection.%目前，变点检测技术已经广泛应用到各个领域。然而，由于实际生产环境的复杂性，变点检测技术中的常用参数方法往往存在一定的局限性。为了克服这些问题，提出一种新的变点检测的非参数方法，通过互联网检索应为首次应用AUC(曲线下面积)对样本数据进行在线变点检测。该方法将变点检测分为2个阶段：预分析阶段，对样本数据进行加窗处理，通过计算窗口中样本数据的AUC值的方式来间接得到其均值和方差；检测阶段，通过假设检验的方法对经过处理后的样本数据进行变点检测。通过实验仿真，可以观察到此算法与常规CUSUM算法相比，具有更好的稳健性，而且对检测多个突变点的情况同样有效。
展开▼
