This paper considers the model-based change-point problem with an unknown abrupt change for large-scale network observation sequences. To avoid the difficulty of calculating the normalization coefficients that let the axioms of probability hold, such as $Z(\mathrm{\Lambda })$ in the Exponential Random Graphical Model (ERGM), we present the measure ratio statistics to replace the likelihood ratio statistics. Since the parameter difference reflecting the abrupt change for the network observation sequences is unknown, we first select the dimensions where the parameter difference exists through the $L_1$-norm penalized maximum likelihood estimation process, then propose a CUSUM multi-chart scheme based on the selected dimensions. Moreover, an optimal design of the CUSUM multi-chart is given when $AR{L}_0$ (in-control Average Run length) is large. Two examples are used to illustrate the related theoretical results.

本文考虑了大规模网络观测序列未知突变的基于模型的变点问题。为了避免计算使概率公理成立的归一化系数的困难，例如$Z(\mathrm{\Lambda })$在指数随机图形模型 (ERGM) 中，我们提出了度量比统计来代替似然比统计。由于网络观测序列反映突变的参数差异是未知的，我们首先通过${升}_1$-norm 惩罚最大似然估计过程，然后根据所选维度提出 CUSUM 多图方案。此外，当给出 CUSUM 多图的优化设计时$\mathrm{一种}\mathrm{电阻}{升}_0$（控制中的平均运行长度）很大。用两个例子来说明相关的理论结果。