Aiming at the simulation modeling barriers and data interaction gaps caused by deep coupling among multiple physical domains in new power systems, a “layer-anchor” digital twin modeling and dynamic inference framework for multi-network coupled systems is proposed. First, “layers” are constructed to encapsulate the topological structure and physical laws of single physical domains, while “anchors” are utilized to explicitly define the constraint equations and interaction logic of cross-domain coupling entities, enabling a unified representation of heterogeneous models at the underlying data structure level. Second, a dual-mode collaborative computing engine integrating optimization-based solution and sequential solution is designed. A global unified time-sequence management mechanism is introduced to ensure precise synchronization and consistency of cross-domain data across multiple time scales. Finally, a co-simulation scenario is constructed based on the PJM 5-bus power system and a high-fidelity synthetic meteorological field, and week-ahead economic dispatch together with continuous power flow simulations are carried out for validation. The results show that the proposed framework can effectively handle the deep coupling between energy flows and value flows, achieve global optimal operation, and enable accurate dynamic state inference driven by meteorology conditions. It solves the semantic interoperability issues among heterogeneous systems and verifies its effectiveness in digital twin modeling of complex coupled systems.