Abstract: Existing thermal-hydraulic system analysis codes usually rely on empirical correlations to calculate key physical model parameters related to heat transfer, mass transfer, and phase change. However, under complex conditions that deviate from the original experimental fitting ranges, such as high pressure, low flow rate, mixed convection, and the presence of non-condensable gases, the predictive accuracy and applicability of conventional empirical correlations may be limited. To improve the predictive capability for key model parameters in system analysis codes, this study proposed a coupling framework between a system analysis code and automated machine learning. In this framework, the system analysis code solved the mass, energy, and momentum conservation equations to obtain thermal-hydraulic state parameters. The automated machine learning model was used to predict key model coefficients required for local physical processes such as heat and mass transfer. And the general-purpose interface program was responsible for invoking both the system analysis code and the automated machine learning model, coordinating the time advancement, and transferring variables among the thermal-hydraulic state parameters, key physical model coefficients, and the input and output variables of the machine learning model. Taking the prediction of condensation heat transfer coefficient under conditions with non-condensable gases as an example, the previously developed and validated AutoGluon model for condensation heat transfer coefficient prediction was selected as the validation object. Through the general-purpose interface program, online coupling between this model and the MOSAP system analysis code was realized. The results show that the proposed coupling framework enables stable and efficient invocation of the system analysis code and the automated machine learning model during transient time-step advancement. This method provides a feasible pathway for model extension and accuracy improvement of system analysis codes under complex accident conditions, and offers methodological support for thermal-hydraulic state prediction, model uncertainty analysis, and accident mitigation strategy evaluation in severe accident management.