With the continuous and rapid growth of motor vehicle ownership, the road traffic flow increases steeply, bringing great pressure to the urban road network, resulting in increasingly serious urban road congestion. Accurate prediction of road traffic flow is the basis for scientific and dynamic control of traffic facilities, reducing congestion and improving operational efficiency and road safety. To capture the spatio-temporal correlation that exists between nodes in a real traffic network, this paper proposes a method for predicting traffic flow based on Graph Sampling Aggregation Network and Long Short-Term Memory (LSTM) network, which learns spatial relationships through Graph Sampling Aggregation Network (GraphSAGE), splices the feature information of each roadway node at each collection moment into a vector as the features of the input LSTM unit, and utilizes the Long Short-Term Memory Network (LSTM) to achieve the temporal relationship learning of traffic flow features, and then the current moment flow features are fused to predict the node flow. To verify the model accuracy, it is compared with other centralized neural network models. The results indicate that the presented GraphSAGE-LSTM model has higher prediction performance than the other three control models, and can meet the requirements of traffic flow prediction based on different time intervals of samples.