Traditional fault analysis and prediction rely on statistical methods that have limitations. They struggle when it comes to capturing the temporal characteristics and long-term dependencies of equipment failures. Currently, a type of fault diagnosis and prediction method applied for power equipment secondary is RNN-based. To begin with, the training samples are the historical states and observations of devices collected in this paper. Then, an RNN model was established to understand and extract the temporal relationships and long-term dependencies with its cyclic structure and memory unit. Thirdly, we capture the temporal patterns and feature representations of equipment failures by use of this knowledge during training. In the end, with new observation data, the RNN model that has already been trained is presented with fault diagnosis and prediction capability. Experiments results suggest that the RNN-based fault diagnosis and prediction method outperform the power equipment secondary. Compared with traditional fault analysis and prediction methods, RNN based fault diagnosis and prediction methods can more accurately capture the time characteristics and long-term dependencies of equipment faults, which can improve the accuracy and precision of prediction.