The secondary circuit involves a large number of electrical parameters and signals, which may contain noise and have complex correlations with each other. Faults are often difficult to directly observe and their temporal characteristics are difficult to extract, making hidden fault detection difficult. To this end, a method for online detection of hidden fault states in the secondary circuit of substation relay protection based on recurrent neural networks is proposed. Construct a secondary circuit state detection framework, use electronic transformers to obtain the working state set information of secondary equipment, and analyze its reliability. Using recurrent neural networks to construct an online detection model for hidden fault states in secondary circuits, recurrent neural networks have significant advantages in processing time series data. Based on this, determine the initial weight value of the online fault state detection model, update the learning factor of the model training, and output the hidden fault state of the secondary circuit. The experimental results show that the F1 scores of the secondary circuit hidden fault state detection under the proposed algorithm are all around 0.981, and the convergence speed of this method is the fastest, indicating that the proposed hidden fault state detection method is highly efficient.