The invention relates to a method and device for modeling a long-time-scale photovoltaic output time series. The method comprises collecting the historical data of a photovoltaic power station, selecting the photovoltaic output with a time length of 1 year and a time resolution of 15 minutes, and obtaining daily weather types corresponding to the photovoltaic output from a meteorological station, and calculating the transition probability among the respective weathers; generating a simulated time series of the photovoltaic output within a preset time scale and verifying the validity of the simulated time series. The method can obtain the annual and monthly simulated time series satisfying a photovoltaic time series random fluctuation rule according to different requirements, and provides favorable conditions and data support for the time series production simulation of large-scale new energy.