Abstract: A high-efficiency, high-accuracy prediction of the thermal field in ablative materials forms the foundation for the effective design and reliability enhancement of thermal protection systems (TPS) for hypersonic vehicles. To address the coupled chemical reactions and pyrolysis heat-conduction problem inherent in ablative materials, we propose a Physics-Informed Neural Network (PINN) framework for rapid prediction of the temperature field. In this framework, spatial coordinate and time serve as model inputs, and a dual-branch architecture is employed in the latent layers to concurrently learn the evolution of material temperature. A composite loss function is constructed from the residuals of the heat-conduction equation for ablative materials, the Arrhenius reaction equation, the initial-condition constraint, and the boundary-condition constraint, and we adjust the weights of various loss terms to introduce an adaptive weight strategy, thereby balancing the various physical constraints. Comparative analysis with numerical results obtained via the Finite Volume Method (FVM) demonstrates that the proposed framework achieves a three-order-of-magnitude enhancement in computational efficiency for predictive simulations, while maintaining prediction errors within a controlled bound of 6%. This approach thus offers a practical and effective means for rapid iteration in the design of ablative thermal protection systems.