Short-term power forecast is an important way to guide operation of renewable energy stations and hybrid energy system (HES). The current studies focus on power forecast of single renewable energy station. However, the universality and applicability of power forecast model for HES is not clear. This study proposes a physics and data dual-driven day-ahead power forecast model for hydro–wind–photovoltaic HES. The WRF model and Xinanjiang model are used to drive meteorological and hydrological forecasts respectively. The hybrid variational mode decomposition - principal component analysis method is applied to further extract the features hidden in the meteorology or hydrology factors. The long short-term memory network is used to drive power forecast. China’s Guandi hydro–wind–PV HES is considered as a case study. Results show that the forecast root mean square error of dual-driven model decreases by 4.2% ~ 12.0% compared to single-driven model.