A new method for generating medium and long-term photovoltaic output scenarios based on generative adversarial networks is proposed to address the problem of high variable dimensions in generating annual photovoltaic output sequence scenarios.Firstly，the fuzzy C-means clustering （FCM） algorithm is used to divide daily meteorological states，and the temporal nature of daily meteorological information is simulated based on generative adversarial networks.Then，a state division network is introduced，and a state generative adversarial network is constructed to simulate the distribution pattern of meteorological states during the week.Photovoltaic output sequence scenarios are generated by layering from weekly and daily scales.Finally，historical meteorological and photovoltaic output data from a photovoltaic power station in China are used to verify the effectiveness and accuracy of the proposed method.