Concrete box-girder bridges subjected to non-uniform temperature fields are prone to thermal stress-induced cracks and deformation, which seriously threaten structural safety and durability. Accurate prediction of the temperature field serves as the critical foundation for real-time monitoring, damage early warning, and preventive mitigation strategies. To address the insufficient accuracy of discrete-point temperature predictions under multi-factor meteorological features, this study proposes a GA-optimized CNN-BiLSTM-Attention temporal model. The model integrated Bidirectional Long Short-Term Memory (BiLSTM) networks to capture bidirectional temporal dependencies in time-series data, and integrated Convolutional Neural Networks (CNN) to extract localized spatial features from time-series data, while further incorporating a self-attention mechanisms to dynamically assign weights to meteorological parameters, thereby enhancing sensitivity to critical features driving temperature variations. Concurrently, GA is employed to optimizes hyperparameters ( the number of hidden units, learning rate, etc.) to improve prediction stability and generalization capability. Validation results based on field monitoring data from Nanchong Jialing River Bridge demonstrated that the model's superior performance in both short-term (S1 segment) and long-term (S2 segment) predictions, achieving RMSE values as low as 0.178 and 0.129 respectively, with the coefficient of determination (R2) approaching 0.99. Compared to baseline models (e.g.,LSTM, BiLSTM), the error reduction exceeds 20%, while its capability to capture temperature peaks exhibited substantial enhancement. This research validated that the hybrid model integrating spatiotemporal feature fusion and dynamic weight allocation effectively could resolve complex meteorological coupling mechanisms, while providing a high-precision temperature field prediction tool for bridge structural health monitoring. This advancement possessed significant engineering implications for enhancing infrastructure service life and advancing safety-oriented operation and maintenance practices.