The efficiency of intelligent substation relay protection test needs improvement. There is also a problem in that a digital relay protection test device cannot automatically diagnose faults during the whole test process. Thus a relay protection test fault diagnosis method based on long short-term memory (LSTM) network is proposed. In this paper, the fault section characteristic information and fault categories are first established. Then a multi-fault diagnosis model is established, and a fault diagnosis process is constructed. Taking the typical 220 kV relay protection test as an example, by comparing LSTM with cyclic neural, a BP and a deep neural networks, the three evaluation criteria (mean absolute error, accuracy and comprehensive evaluation index) of the LSTM model are found to be optimal when the actual fault information and partial unreliable information are two different inputs. Simulation results show that the relay protection test fault diagnosis method based on the LSTM network has high precision and good fault-tolerant performance. This work is supported by the National Natural Science Foundation of China (No. 11872069).