Abstract: Rapid prediction of compressible flows based on parametric approaches is critical for aerodynamic design and optimization in aerospace engineering. While physics-informed neural networks (PINNs) provide a novel paradigm for solving such problems, they face significant challenges in handling flows with shock waves, specifically regarding shock capturing failure and numerical oscillations. To address these issues, this paper proposes a shock detector-guided adaptive viscosity PINN method. First, a “detection-guidance-penalization” adaptive viscosity control mechanism is established. By leveraging a continuous shock indicator function generated by a shock detector, this mechanism realizes adaptive dissipation augmentation in shock regions to enhance stability, while intensifying penalization in smooth regions to minimize numerical dissipation. Furthermore, a Mach number-layered and loss-guided adaptive resampling strategy is introduced to resolve learning imbalances within the parameter space, implemented alongside a four-stage progressive training framework. Finally, the prediction accuracy is validated using single-wedge and three-wedge inlet benchmark cases. The results demonstrate that the proposed method, which requires no external data, effectively overcomes the accuracy and stability issues inherent in existing PINN methods for solving strong discontinuity flow problems. In the single-wedge case within the parameter domain of Mach 2.0 to 3.0, the shock angle prediction error remains below 0.2^°. For the three-wedge inlet case, the mean absolute error of the physical fields is controlled within 1.0%, and the shock angle prediction error is below 0.3^°. The method accurately captures shock strengths and flow field structures under various conditions, verifying its effectiveness and robustness in solving parameterized strong discontinuity flow problems.