To address the issue of typhoon-induced rainstorm-flood secondary disasters in power grids

this study proposes an outage mechanism-data collaborative spatiotemporal prediction model to calculate the outage probability of power distribution transformers under rainstorm-flood disasters. First

a hydrodynamic model (InfoWorks ICM) is employed to construct an urban flood simulation system

calculating transformers inundation depths for preliminary outage probability assessment. Second

a mechanistic flooding model is integrated into a multilayer perceptron by con-straining its loss function

and hydrological mechanisms are embedded into data-driven training to establish the hybrid mechanism-data collaborative model. Third

in the model

the power grid and geographic and meteorological data are synthesized to achieve spatiotemporally outage predictions. Finally

an empirical analysis is conducted on the distribution trans-formers of a city’s power grid in Guangdong Province during Super Typhoon Yagi (2024). Results demonstrate that the proposed framework significantly improves prediction accuracy while reducing deviations from theoretical mechanisms.