In the dual carbon background, the large-scale increase of distributed photovoltaic power generation and grid-connected access have brought great challenges to the new power system. Due to the influence of weather factors, the high permeability distributed photovoltaic output and power load have strong uncertainty and volatility, which to a certain extent increases the difficulty of predicting the net power of distribution network. In order to improve the prediction accuracy of net power of distribution network, this paper puts forward the net power prediction method of attention-Bidirectional GRU neural network. Firstly, the paper analyzes the photovoltaic output characteristics, user-side load characteristics and the influencing factors of the net power of distribution network, and fully grasps the influence of the net power on the variation law of distributed photovoltaic output and user-side load. Then the net power prediction model of distribution network is established by integrating Attention mechanism into Bidirectional GRU neural network. Among them, The Attention mechanism gives different Attention to the input features, and the Bidirectional GRU neural network can learn the temporal characteristics of net power. The perfect combination of the two greatly improves the representation and generalization ability of net power prediction model. Experimental results show that the proposed method greatly improves the accuracy of net power prediction of