Accurately predicting bike-sharing flow is essential for optimizing the supply-demand balance of shared bikes and enhancing urban residents' travel convenience. To address the issues of low prediction accuracy and insufficient capture of spatiotemporal characteristics in bike-sharing flow prediction, a hybrid convolutional-recurrent neural network (Conv3D-GRU) model was proposed. Using Chicago's 2022 full-year bike-sharing data, experiments were conducted, and the results were compared with those of the 3D convolutional neural network (3D-CNN) model and the convolutional long short-term memory (ConvLSTM) model. The model performance was evaluated using root mean squared error (RMSE), mean absolute error (MAE), and the coefficient of determination(R²). Experimental results show that compared with the 3D-CNN and ConvLSTM models, Conv3D-GRU is improved by 3.25%, 4.90%, 1.14% and 11.94%, 13.70% and 2.46% on RMSE, MAE and R², respectively. This demonstrates that the Conv3D-GRU model has lower prediction errors and higher prediction accuracy, making it an effective and reliable approach for forecasting bike-sharing inflow and outflow.