Microgrid fault detection is challenged by diverse operating modes, nonlinear system dynamics, and limited availability of labeled fault data. Conventional threshold‐based and model‐based methods often exhibit reduced robustness under noise and multi‐parameter fault conditions, while data‐driven machine learning approaches typically require extensive training data. This paper proposes a Deep Transfer Convolutional Neural Network (DTCNN) framework for microgrid fault detection, leveraging a pretrained VGG‐16 model to enhance feature extraction under limited data conditions. Voltage, current, and frequency signals are transformed into time–frequency spectrograms using the Short‐Time Fourier Transform (STFT), enabling effective representation of transient and steady‐state fault characteristics. Simulation results based on an industrial park microgrid demonstrate that the proposed DTCNN consistently outperforms conventional machine learning methods, including Support Vector Machines and Random Forests, in terms of accuracy, recall, and F1 score. Moreover, the proposed framework maintains stable detection performance across different operating modes and load conditions, indicating strong robustness and generalization capability. These results suggest that deep transfer learning provides an effective and practical solution for reliable microgrid fault detection. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.
Su et al. (Sun,) studied this question.