Abstract: The problem of estimating surface heat fluxes is a typical inverse heat conduction problem. To solve this problem, the estimation of surface heat fluxes based on physics-informed neural network(PINN) is proposed. Firstly, the fully connection neural network satisfying the heat conduction problem’s input and output is constructed. And then, the residual error of one-dimensional heat conduction equation is integrated into the loss function of the neural network as a physical knowledge constraint by using automatic differentiation technology. Finally, the neural network is trained by the temperature data provided by the direct heat conduction problem. When the loss function converges sufficiently small, the neural network model can approximate the heat conduction equation, and then the neural network model can be used to identify the surface heat flow. The results show that the estimation of surface heat flux based on PINN can effectively estimate the surface heat flux, especially under the condition of sufficient information, the results are satisfactory, which verifies the effectiveness of the method. Compared with conventional identification method, the PINN method integrates data-driven modeling with physical equation constraints, thereby achieving more accurate identification results. When using noisy data to train the model, the anti-noise ability of the model can be improved by reducing the weight of temperature data, the method based on PINN is robust to noisy data.