With the continuous integration of renewable energy sources and power electronic devices, the disturbance resistance of power systems has gradually declined, increasing the risk of frequency instability. Therefore, predicting the frequency response under anticipated faults in steady-state operation, assessing frequency security risks, and taking timely measures to ensure system frequency security are of great importance. However, existing methods face a trade-off between prediction accuracy and computational speed. To address this issue, this paper proposes a frequency response prediction method for power systems under anticipated faults based on a Physics-Guided Convolutional Neural Network (PG-CNN). The proposed method takes steady-state power flow information and anticipated fault data as input features and incorporates the swing equation to design physics-guided terms. A physics-guided loss function is constructed to guide model training, thereby enhancing the interpretability of the deep learning model. Simulation tests conducted on an improved IEEE 39-bus system demonstrate that, compared to purely data-driven methods, the proposed approach can rapidly and accurately predict the frequency response curve of the power system under anticipated faults. Furthermore, it maintains high prediction accuracy even in small-sample scenarios.