With the construction of the new power system, distributed photovoltaic will be used in a higher proportion, so the problems of voltage fluctuation and abnormal node state of the distribution network will become more and more prominent. The traditional analysis method based on power current has limited detection capability in topological incompleteness and dynamic environment, which is difficult to support the monitoring and abnormal identification of distribution network operating status. To solve this problem, this paper proposes a distribution network voltage and three-phase unbalance detection method based on the combination of graph neural network (GCN) and long-short-term memory network (LSTM), taking the IEEE33-node distribution network as an example. The experimental results show that the method maintains high detection accuracy under dynamic disturbance scenarios and incomplete topological information conditions, providing a new solution for voltage fluctuation and three-phase unbalance anomaly identification and state assessment of distribution networks.