Abstract: To address the problem of low accuracy for distributed photovoltaic power forecasting due to the general lack of meteorological monitoring devices, this paper proposes an ultra-short-term prediction of distributed photovoltaic power method based on an adaptive graph convolution recurrent network that can realize accurate power output prediction in the absence of meteorological data. Firstly, it is analyzed that the photovoltaic output data have both temporal and spatial correlation. Secondly, the temporal correlations are extracted by gated recurrent unit and an adaptive graph convolution network is utilized to mine potential spatial correlations that traditional graph convolution networks can't capture. Thirdly, an adaptive graph convolution recurrent network is proposed to extract the temporal and spatial correlations of multiple distributed photovoltaics by combining the adaptive graph convolution network and the gated recurrent unit, and the attention mechanism is used to assign weights to the spatio-temporal characteristics at different time. Finally, the final prediction result is output through the fully connected layer. Compared with other methods in different forecasting horizons by using the actual photovoltaic output data, the results show that, compared to traditional gate recurrent network, the average absolute error of the proposed method is reduced by 16.9%, 19.8%, and 30.5% when the forecasting horizon is 15 minutes, 30 minutes, and 60 minutes, respectively.