Abstract: A method for solving the forward and inverse problems of the heat conduction equation based on a physics-informed neural network (PINN) is proposed. In this method, the one-dimensional heat conduction equation is integrated into a loss function as the regularization term using the automatic differentiation. Consequently, the solution of the equation as well as the embedded unknown physical parameters can both be obtained by minimizing the loss function of the PINN. The convergence errors of solving the forward problem and the robustness of parameter identification are analyzed. Results show that for a given network structure, the convergence errors of solving the one-dimensional heat conduction equation based on PINN are is mainly determined by the sampling errors when sampling points are insufficient, while for abundant samples, the convergence error is determined by the optimization errors. Since the loss function contains equation-related regularization terms and adopts the automatic differentiation technique, the identification method based on PINN is robust to noisy data.