Abstract: Using graph neural network-based intelligent analysis as the core technique, this paper investigates knowledge graph-based risk modeling for power-sector application scenarios. It proposes a precise method for identifying the impact scope of information system operations by integrating a knowledge graph with a dynamic time-series monitoring algorithm. The core innovations include multi-entity operational risk knowledge graph modeling, knowledge graph-based identification of operational impact scopes, and message-passing-based operational risk prediction. The method integrates the static asset topology of the configuration management database (CMDB), time-series metrics from real-time performance monitoring, and end-to-end dynamic call data from application performance monitoring (APM) to construct a multi-entity relational risk knowledge graph. It further introduces a multidimensional quantitative evaluation model for deviations from system stability and a risk propagation inference algorithm for service call chains. These components enable numerical calculation of information system operational risks, end-to-end traceability, and graded quantitative assessment. Field test results show that the proposed technology effectively enhances intelligent operation and maintenance and proactive security risk prevention and control in power information systems. It provides methodological and engineering support for the long-term reliable and stable operation of new power systems.