This paper proposes a novel short-term forecasting method for purchased photovoltaic (PV) generation. The proposed method is used to solve emerging problems, such as low accuracy of electricity load forecasting, which are associated with the rapid increase in PV generation. In the present study, hourly PV power is first modeled in the form of state-space models (SSMs), which incorporate a local power model and PV system parameters. Hourly installed PV capacities are then estimated using data that are available on a monthly basis. Finally, using the hourly capacities and weather observations, data assimilation in the SSMs is performed by an ensemble Kalman filter. As a result, the hourly physics-based PV power models are enhanced by monthly PV purchase volumes and significantly outperform an existing operational model. Furthermore, it is possible to simultaneously estimate PV system parameters, such as the coefficient of PV conversion, in the data-assimilation process. (C) 2017 Elsevier Ltd. All rights reserved.