To address the issue of power system failures and power outages caused by the flooding of power facilities during rainstorm, this paper proposes a dynamic assessment and early warning mechanism based on a two-dimensional (2D) hydrodynamic model and data fusion. First, by combining the Chicago rainfall model with a 2D hydrodynamic model, the depth of water accumulation at key power facility locations is accurately simulated, significantly improving prediction accuracy. Next, a comprehensive evaluation system is developed to quantify the risk of power facility failures due to flooding, taking into account factors such as water depth, equipment type, and potential failure modes. In addition, a short-term load forecasting model (VMD-LSTM) integrating variational mode decomposition (VMD) and long short- term memory (LSTM) is constructed to provide accurate forecasting of grid load changes during rainstorm as an critical input for risk assessment. Finally, a risk warning level scheme is designed to accurately identify the power grid nodes requiring warning, providing scientific decision-making support for operators. Through the analysis of specific regional power grid cases, the effectiveness and practicality of the propsoed method are demonstrated, providing solid technical support for disaster prevention and mitigation.