【目的】电力系统数据表现出高随机、强互动、多耦合等特征，传统基于条件假设和模型简化的机理建模仿真分析其时效性无法满足大电网实时防御要求。利用大电网广域时空序列信息可以提高分析的精确性和时效性，突破机理模型的限制，对电网物理运行状态进行多层次态势感知。【方法】采用数据驱动并结合电网动态特性，基于随机矩阵理论对电网时空数据进行分析挖掘，克服传统基于电网物理拓扑的局限，实现关键节点的实时评估和辨识；根据关键节点评估结果，基于数据驱动采用马尔科夫链，提出扰动传播预测方法，实现对电网节点个性化风险评估和电网态势群体性预测分析。【结果】通过个性化风险评估电网节点，找出系统中风险节点即关键节点，探索将监测预警向预测预警转变；利用马尔科夫链预测相对受扰节点从而预测扰动传播，通过算例分析验证所提方法的有效性和合理性。该方法可在电力系统发生扰动后，对即将引发系统严重事故的关键节点进行重点监控，相比于其他辨识方法，该方法既考虑了系统电气特性，又考虑了系统拓扑特性，辨识结果会随着系统运行状态的改变而改变。【结论】通过提前预防控制保证系统的安全稳定运行，对防御大停电事故具有重要意义。

[Objective] Power system data demonstrates features of high stochasticity， strong interaction， and multi-dimensional coupling. Traditional mechanistic modeling and simulation analyses， which rely on conditional assumptions and model simplifications， fail to meet the real-time defense requirements of large-scale power grids in terms of timeliness. By fully leveraging wide-area spatiotemporal sequence information from power grids， we can enhance analysis accuracy and timeliness while transcending the limitations of mechanistic models. This approach enables multi-level situational awareness of the physical operational states of power systems， facilitating more effective real-time defense strategies. [Methods] This paper presents a method to promote the safe and stable operation of the power grid. Based on the dynamic characteristics of the power grid， the spatiotemporal data are analyzed and excavated by random matrix theory. This method overcomes the limitations of the traditional grid-based physical topology， realizing the real-time assessment and identification of critical nodes in the system. According to assessment results of critical nodes， the disturbance propagation is predicted based on data-driven Markov chain. The method achieves the personalized risk assessment of grid nodes and collective predictive analysis of grid situations. [Results] This paper focuses on personalized risk assessment of nodes in the power grid， identifying risk nodes or key nodes in the system. By using the method proposed in this paper to explore the transition from monitoring and warning to predictive warning. The effectiveness and rationality of the proposed method can be verified through case analysis by using Markov chain to predict the propagation of disturbances relative to disturbed nodes. This study enables early warning of the disturbance propagation when the power grid disturbance occurs， and focuses on the critical nodes that will cause serious accidents in advance. Compared with other identification methods， the proposed approach takes into account both the electrical characteristics and topological properties of the system， with its identification results dynamically adapting to changes in system operating conditions. [Conclusions] Prevention and control are conducted in advance to ensure the safe and stable operation of the system， which is of great significance to the defense of blackouts.