This paper constructs a hybrid data-driven and physics-based model for temperature prediction of automotive exterior components based on insufficient natural exposure experimental data. The hybrid model consists of the physical part on heat conduction and the data-driven part on artificial intelligence algorithms. The physical part is developed by the heat balance equation, which ensures the generalization ability of the model. The data-driven part is developed based on BP neural network to calculate the unknown parameters and estimate the missing weather data, which are necessary for the physical part. At the end of the paper, the model performance is evaluated by the experiment data. The numerical results illustrate that the hybrid model not only has accurate predictive ability, but also possesses good generalization capability.