Conventional reservoir engineering and numerical simulation methods have drawbacks such as complex data requirements and limited applicability in production index prediction. To address these issues, a prediction method for oil well development indicators based on T-GNN (timing graph neural network) and VCA (variable correlation attention) model was proposed. The injection-production well pattern was abstracted as a graph structure. The T-GNN was employed to quantitatively characterize and screen the inter-well injection-production dynamic response relationships at different time steps. Considering the temporal sequence characteristics of oil well development indicators and injection well water injection indicators, the VCA model was used to predict oil production rate and water cut. A case study in the M-1 Oilfield demonstrates that the root mean square error and determination coefficient for oil production prediction are 9.18 and 0.92, respectively, while those for water cut prediction are 2.63 and 0.89, respectively. The proposed method can accurately predict oil production rate and water cut, which is of great significance for promoting the efficient application of intelligent algorithms in oil well development indicator prediction.