In order to address the lack of systematic theoretical guidance and the difficulty in effectively representing system control-feedback mechanisms in Bayesian network construction for dynamic risk assessment, a risk assessment method integrating System-Theoretic Accident Model and Processes (STAMP) with Dynamic Bayesian Network (DBN) was proposed. First, a risk assessment indicator system was constructed by identifying Unsafe Control Actions (UCAs) based on safety control structures. Then, three mapping rules for the transformation of DBN topological structures were established to realize the logical transition from qualitative risk analysis to quantitative assessment models. Furthermore, the bidirectional reasoning function was utilized for risk prediction and identification of key risk factors. Finally, empirical analysis and validation were conducted using the US-Accidents dataset. The results demonstrate that the method accurately captures the stage-wise evolutionary characteristics of key risk factors, enables dynamic risk propagation analysis as well as key risk factor identification, and reveals the influence of system control-feedback loops on risk evolution. It is concluded that the proposed method achieves the accurate transformation of assessment models from qualitative analysis to quantitative assessment through the triple mapping mechanism, providing a structured modeling basis for the dynamic risk assessment of diverse systems.