Cross-Domain Energy Flow Digital Twin: A Dynamic-Symbiosis-Driven Framework for Multi-Energy-Flow Coordinated Regulation in Low-Carbon Systems Based on Three-Dimensional Energy Spatio-Temporal Graph

The transition toward Low-Carbon Integrated Energy Systems (LCIES) is pivotal for achieving global carbon neutrality; however, coordinating multi-energy flows (electricity, heat, and hydrogen) faces formidable challenges validated by recent high-impact research: insufficient modeling fidelity, the disconnection between data and decision-making, and inconsistent multi-objective regulation. To overcome these bottlenecks, this study proposes a Cross-Domain Energy Flow Digital Twin (CDEFT) grounded in a “Dynamic Symbiosis” paradigm, based on a three-dimensional energy spatiotemporal graph. First, we construct a “physical-virtual” closed-loop framework that integrates mechanism models with dynamic graph models, utilizing a two-level synchronization mechanism to achieve real-time perception. Second, to bridge the gap between “data mapping” and “decision optimization,” we design a hierarchical decision mechanism based on the Hierarchical Markov Decision Process (HMDP). In this architecture, Spatio-Temporal Graph Neural Networks (STGNN) extract global state embeddings from structured “geography-physics-temporal” graph sequences, which then guide the Hierarchical Reinforcement Learning (HRL) agent to balance macro-strategies (carbon-safety-benefit) with device-level precise control. This work establishes a novel STGNN-HRL fusion paradigm, which, to our knowledge, is the first closed-loop framework to integrate dynamic spatio-temporal graphs, physics-informed STGNN, and STGNN-driven hierarchical decision-making for electricity-heat‑hydrogen systems. Validated across five dimensions, the proposed CDEFT demonstrates significant performance improvements, strictly forming a “mechanism-performance-value” logical chain. Notably, experimental results show that the method achieves a 27.1% carbon emission reduction rate and shortens fault recovery time to within 2.1 min under extreme scenarios. Furthermore, in real-world engineering validation, the framework maintains a cost saving rate of at least 14.3% under extreme weather conditions, with an electricity-heat‑hydrogen collaborative contribution degree reaching 82.3%, effectively suppressing cross-domain fault propagation by 45.6%. These results validate the Dynamic Symbiosis paradigm and provide a robust and scientific technical pathway for the safe, low-carbon operation of integrated energy systems.