<p>The effective integration of heterogeneous data from SCADA (Supervisory Control And Data Acquisition), PMUs (Phasor Measurement Unit), weather sensors, and dispatcher logs is critical for situational awareness and rapid decision-making in modern power grid control centres. It is evident that challenges such as information overload and the semantic gap between physical measurements and operational directives often impede response times during critical events, such as fault diagnosis and disaster recovery. The proposed integrated framework integrates STURM (Spatio-Temporal Unified Representation Model) with HASM (Hierarchical Attention Summarization Mechanism), herein referred to as STURM-HASM. The STURM layer constructs a dynamic heterogeneous graph to unify representations of physical grid assets, real-time operational parameters, and dispatcher commands, capturing their spatio-temporal dynamics. The HASM layer incorporates a dual-channel generator, with a rule-based channel that ensures strict adherence to power system safety standards, and a neural channel that generates fluent natural language summaries using a hierarchical attention mechanism. The dynamic fusion module under discussion is capable of adapting operational safety constraints in such a manner as to achieve a balanced state of semantic fluency. Extensive evaluations on power system-specific datasets, including a custom PowerGrid-NET, demonstrate the framework's efficacy. The system has been demonstrated to achieve 95.1% accuracy in the identification of equipment anomalies, a summary quality (ROUGE-L) of 0.83, and a 40% improvement in the recall of key information. It is imperative to note that the system exhibits a response latency of less than 200&#xa0;ms, thereby satisfying the real-time requirements of grid dispatch. A digital twin simulation at the provincial level demonstrated a 50% reduction in fault analysis time, thereby significantly enhancing power system emergency management.</p>