With the refinement and intelligentization of power grid optimization and the extensive adoption of advanced applications of power grid security and economy, and the large-scale access of distributed energy, the accuracy requirements of bus load forecasting are constantly increasing while uncertainty and nonlinear of the load are further enhanced. Aiming at the above problems, an ultra-short-term bus net load forecasting model based on phase space reconstruction and deep belief network is proposed in this paper.firstly, the phaes space reconstruction of the original time series is carried out by C-C method, and then the reconstructed data is fitted by the deep belief network to obtain the predicted value of the load. In this paper, the effectiveness of the proposed ultra-short-term bus load forecasting model is tested by using the measured load data of a substation in a city. It is proved that the proposed model still has high prediction accuracy under the condition of high distributed power penetration rate and large fluctuation of bus load.