Abstract: Aiming at the problems of uneven measurement and insufficient accuracy of measurement in distribution networks, a super-resolution measurement generation method based on graph attention networks is proposed. The method can improve the spatio-temporal resolution of distribution network states, and has the ability of topology generalization. It can adapt to the topology reconfiguration conditions of distribution network to minimize measurement acquisition and achieve high-precision state awareness of distribution networks. The proposed method uses the attention mechanism to learn the correlation between electrical state variables of adjacent buses, adds power flow constraints to model training, and designs the minimum topology set required for model training, so as to avoid overfitting training sample data and improve the generalization ability of the proposed model. The effectiveness of the proposed method is verified by IEEE 33-bus and IEEE 123-bus standard distribution network cases.