Power grid system is a critical infrastructure and traditional detection methods are difficult to deal with its dynamic assets, complex business logic, and high concealment attacks (such as fixed value tampering). A model driven multi security constraint detection framework is proposed to address the issues of asset perception ambiguity and low threat detection efficiency in the dynamic monitoring system of the power monitoring system. Firstly, construct a tripartite model of network security space (asset model, relationship model, behavior model) to achieve precise and topological perception of business system assets; Secondly, integrate the Adversarial Tactics, Techniques, and Common Knowledge (ATT&CK) attack chain with the characteristics of power grid business, and design a power grid specific attack knowledge graph; Finally, by combining graph constrained inference with behavior sequence analysis, a Multi-Constrained Threat Detection algorithm (MC Detector) is proposed. Experiments have shown that this method achieves an asset recognition accuracy of 98.7% in real power environments, reduces the false alarm rate of threat detection to 1.2%, and increases response speed by three times.