Abstract: With the advancement of machine learning techniques, data-driven flow field prediction has become a significant research focus in fluid mechanics. Compared with traditional neural networks, which often require large amounts of training data and suffer from poor interpretability, physics-informed neural networks (PINNs) embed physical constraints directly into the loss function. This enables accurate flow field predictions with only minimal training data. This study investigates the reconstruction of shear flow around a circular cylinder under low Reynolds number conditions using a PINN-based approach. Numerical simulation data generated by the open-source computational fluid dynamics (CFD) software OpenFOAM were used to construct the training dataset. A PINN model incorporating shear flow boundary conditions for circular cylinder flow was developed, and velocity and pressure information from sampling points within the flow field were employed to train the model. The goal was to predict the velocity and pressure distributions across various regions of the flow field. To evaluate the performance of the PINN model, its predictions were compared with results from numerical simulations, demonstrating the model's ability to effectively reconstruct complex flow fields. Furthermore, this study investigated the impact of neural network parameters, such as the number of layers, the number of nodes per layer, and the spatial location of monitoring points, on the prediction accuracy. The mechanisms underlying the optimization of these parameters were analyzed in detail. The results indicate that the PINN model can accurately reconstruct the velocity and pressure distributions of shear flow around a circular cylinder, achieving a prediction accuracy that closely aligns with numerical simulation results. This validates the practical applicability and accuracy of PINNs for reconstructing flow fields in low Reynolds number shear flows. Moreover, comparative analyses revealed that the location of monitoring points has a significant influence on the prediction results. Proper selection of monitoring points can enhance the prediction accuracy by 1 to 2 orders of magnitude.