The transportation department relies on accurate traffic forecasting for effective decision-making. However, determining relevant parameters for existing traffic flow prediction models poses challenges. To address this issue, this study proposes a hybrid model, sparrow search algorithm-gated recurrent unit-long short-term memory (SSA-GRU-LSTM), which leverages the SSA to optimize the GRUs and LSTM networks. The SSA is employed to identify the optimal parameters for the GRU-LSTM model, mitigating their impact on prediction accuracy. This model integrates the predictive efficiency of the GRU, LSTM’s capability in temporal data analysis, and the fast convergence and global search attributes of the SSA. Comprehensive experiments are conducted to validate the proposed method via traffic flow datasets, and the results are compared with those of baseline models. The numerical results demonstrate the superior performance of the SSA-GRU-LSTM model. Compared with the baselines, the proposed model results in reductions in the root mean square error (RMSE) of 4.632%–45.206%, the mean absolute error (MAE) of 2.608%–53.327%, the mean absolute percentage error (MAPE) of 1.324%–13.723%, and an increase in $R^2$ of 0.5%–17.5%. Consequently, the SSA-GRU-LSTM model has high prediction accuracy and measurement stability.