The invention discloses a deep learning-based transient stability distinguishing system and method of a power system, and belongs to the technical field of dynamic safety estimation of the power system. The system comprises four modules of sample acquisition, characteristic extraction, sample expansion and stability distinguishing. The invention simultaneously discloses a method for implementing the system. The characteristic data of sample data of the power system is extracted by a depth auto-encoder, the dimension of a sample is reduced, and the calculation and the calculation time can be effectively reduced; a confrontation generation network is used for generating pseudo sample data, sample set scale is expanded, so that the stability distinguishing module can be used for distinguishing transient stability of the power system very well, and the stability distinguishing accuracy is improved; and a depth convolution neural network is used for building the transient stability distinguishing module of the power system, and the transient stability of the power system can be more accurately distinguished in real time.