Abstract: To address the issues of large alignment prediction errors in finite element models and the strong dependence of traditional data-driven methods on extensive measured data during the construction stage of long-span continuous girder bridges, a bridge camber prediction method based on a Physics-Informed Neural Network (PINN) is proposed. In this method, the Euler-Bernoulli beam theory is incorporated into the neural network's loss function as a physical constraint during the model training process. Combined with sparse monitoring data from the construction stage, the model effectively captures the structural stress and deformation characteristics. Furthermore, the Ivy algorithm (IVYA) is utilized to optimize the network's hyperparameters, thereby improving the training efficiency and prediction performance. Numerical experiments demonstrate that the PINN model exhibits higher accuracy and stability in predicting the cantilever end deflection compared to the traditional GM-BP model. The average relative error is reduced from 6.37% to 0.68%, and the model training error is decreased from 4.136×10⁻² to 1.617×10⁻². These results indicate that the proposed method maintains excellent physical consistency, as well as robust generalization and extrapolation capabilities under sparse data conditions. Ultimately, this method effectively improves the accuracy of pre-camber prediction during bridge construction, providing a highly efficient and reliable new approach for construction alignment control and intelligent monitoring.