With the increase of renewable energy's penetration rate in power systems， it is essential to make real-time， accurate and highly fault-tolerant state estimations for distribution networks to cope with the intermittency of renewable energy and keep safe operation of the power grid. Since the assembly level of distribution network measurement devices is incomplete and the model-driven state estimation can hardly adapt to the high uncertainty of environment， the fused measurement data from SCADA/PMU is adopted to train the gated graph neural network （GGNN）.Then， a high fault-tolerant distribution network state estimation method based on GGNN is proposed. It can obtain the spatio-temporal relationship between the measurement and the state estimation by using the graph convolutional layer and GRU-like to extract high-dimensional spatio-temporal features of the measurement. The proposed algorithmic solution is assessed and validated based on an IEEE 33-bus system and an IEEE 118-bus system， respectively. The assessment result show that GGNN can effectively fit the space-time mapping of measurement and state data with a higher accuracy and robustness compared with the traditional Weighted Least Squares （WLS） and Multi-layer Perceptron （MLP）.
