多变点检测问题的Shape-based BS算法

BS算法是时间序列多变点检测中最经典的算法之一,但是基于全局CUSUM统计量的识别过程会带来过多误判和较高的时间复杂度.BS算法是一种离线的序贯方法,因此没有充分利用数据的时序信息;另一方面,BS算法识别变点的原则是CUSUM统计量最大化,也没有考虑统计量构成序列的形态特性.鉴于此,提出一种基于局部形态识别的BS改进算法,命名为Shape-based BS算法.基于局部形态识别统计量,不仅大大降低计算复杂度,且降低了因变点间的互相干扰而带来的误判率,进而提升变点识别的稳健性.最后,将此算法应用到了电力系统的"场景压缩"问题上,具有满意的实用效果.

Shape-based BS algorithm for multiple change-points detection

BS algorithm is one of the classical algorithms for multiple change-points detection,it may bring about too many misjudgments and a high time complexity due to the procedure of global CUSUM statistic. On one hand,the BS algorithm is an off-line sequential method, therefore the data timing information is not fully utilized. On the other hand, the principle of the BS algorithm to detect the change-points is to maximize the CUSUM statistic,which does not consider the morphological characteristics of the statistical constituent sequence. In view of these, the paper proposes an improved BS algorithm, named Shape-based BS algorithm, which is based on local shape recognition. Basing on the local pattern recognition of statistic sequence not only decreases the computational complexity,but also avoids mutual interference among change-points, and it could also promote the robustness in discerning change points. At last, this paper uses Shape-based BS algorithm to reduce the scenarios of electric power, and achieves satisfactory practical results.