Fiber optic current transformers are important equipment in DC converter stations, and their reliable operation is crucial for the overall safety of the system. Therefore, to conduct comprehensive condition monitoring and life prediction for such equipment is of great significance. A CNN-GRU-based prediction model is proposed to address the difficulty of accurately predicting the state of fiber optic current transformers, which arises from multiple factors. First, the sliding window method is used to construct continuous feature maps of historical data such as outside temperature, driving current of the light source, and driving voltage of modulator as inputs. Secondly, the advantage of using Convolutional Neural Network (CNN) for data feature extraction and dimensionality reduction is used to extract features from input data and obtain feature vectors. Then, input the feature vectors into the Gated Recurrent Unit (GRU) network for prediction. Finally, calculate the daily variation of the driving current of the light source and the driving voltage of the modulator based on the predicted results, and conduct a life analysis. Using operational historical data from the Guquan converter station as an example, and comparing it with GRU and CNN-LSTM models, the results indicate that the proposed method offers better predictions.