The aim of this study is to achieve efficient fault diagnosis for circuit breakers. To address the issue of imbalanced data collection, the Modified Synthetic Minority Over-sampling Technique (MSMOTE) algorithm based on Mahalanobis distance is employed for data augmentation. Additionally, the Firefly Algorithm (FA) is utilized to optimize the number of nodes and learning rate in the Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) hidden layers. The data expanded by the MSMOTE algorithm is input into the FA-CNN-LSTM model for training and classification. Experimental results indicate that the proposed method can efficiently diagnose circuit breaker faults even in scenarios with limited fault samples. With the optimization by the FA algorithm, the classification accuracy reaches 99%. Therefore, the circuit breaker fault diagnosis method proposed in this study exhibits excellent performance, offering a novel and effective approach for the analysis of power grid equipment status.