Dynamic Probabilistic Risk Assessment (DPRA) provides a framework for analyzing time-dependent accident progression and coupled system behavior in nuclear power plants. However, the need to explore large scenario and uncertainty spaces leads to significant computational demands, limiting the scalability of DPRA. Methods that efficiently identify risk-significant conditions and focus analysis on these could reduce this burden, but developing such approaches remains a key challenge. This paper presents a first step toward addressing this challenge by proposing a model-based methodology for guided DPRA. To support this objective, an executable system architecture model is developed using Model-Based Systems Engineering (MBSE). The model captures system requirements, structure, functions, and time-dependent behavior while embedding qualitative risk information such as failure modes and fault logic. Low-fidelity simulations performed on the resulting model enable the identification of risk-significant accident scenarios and parameter combinations that can guide the application of high-fidelity DPRA simulations. The proposed methodology establishes a foundation for improving the efficiency and scalability of DPRA while maintaining traceability to evolving system definitions and requirements. The methodology is demonstrated through a proof-of-concept case study of a representative safety system, illustrating how the architecture model supports guided DPRA simulation and risk-informed design. Future work will integrate high-fidelity simulations by using the scenario insights generated in this work to guide more detailed DPRA.