Composite has advantages such as high specific strength, high specific stiffness, and strong designability and has been widely used in the structure design of modern unmanned systems. However, the traditional finite element method (FEM) has high computational. Data-driven surrogate models also require many samples and the influence of mechanical properties is ignored during the modeling process. The Physics-Informed Neural Networks (PINNs) further introduce real physical constraints based on data-driven surrogate models, which helps reduce the sample number and improve analysis efficiency. This paper provides a surrogate model of composite mechanics based on a data/physics hybrid-driven method. Firstly, the analytical solution of a simple composite laminate is derived and FEM verifies the results. The ply thickness and center deformation of the laminate are used as the sample input and output, respectively. Then different numbers of samples are selected in the same sample space through Latin hyper-cube sampling and analyzed using FEM. Finally, the Back Propagation (BP) Neural Network model of the composite laminate and PINNs model are constructed, respectively. The BP model is only based on the sample data, while the PINNs model additionally contains the static constraint on composite laminate. The results show that the PINNs method only requires a small number of samples to achieve the same accuracy as the BP method, which helps improve the design efficiency. The proposed method can provide a reference for the practical application of artificial intelligence in unmanned systems.