Abstract: This article proposes a theory for the slope stability limit analysis based on physical information neural networks. Based on the lower bound theorem of limit analysis, a neural network is used to construct a self-balancing stress field for slopes. With the help of deep learning techniques, the slope stability control equation is transformed into a neural network loss function, and a limit load solution format is established. Finally, combined with transfer learning algorithms, a fast iterative solution method for the slope stability safety factor is obtained. Through the verification of homogeneous slope examples and of classical theoretical solution examples, the safety factor calculated by this method has an error of less than 9%, compared to the theoretical value, and reduces the calculation time by about 40%, compared to the finite element method. This theory demonstrates good applicability and efficiency under complex boundary conditions.