Aiming at the problems of complex procedures and low computational efficiency of traditional inversion methods of concrete thermal parameters, an inversion method based on physics-informed neural networks (PINN) is proposed. In this method, the learning rate for optimizing the concrete thermal parameters to be inverted is determined according to the accuracy requirements of the parameters to be inverted. Based on the PINN inversion method, the thermal parameters of concrete without and with water pipe cooling were inverted through simulation tests, and the errors of the inversion results were analyzed. The PINN inversion method was compared with the inversion method of traditional finite element calculation combined with genetic algorithm. The results show that the PINN-based method for inversion of concrete thermal parameters has high accuracy, robustness and generalization ability. Using the improved method to determine the learning rate based on the accuracy requirements of the parameters to be inverted can effectively improve the calculation efficiency while ensuring the inversion accuracy. Compared with the traditional inversion method, the PINN inversion method has advantages in terms of a simpler computational framework and higher computational efficiency.