Various faults could happen to the high voltage circuit breaker (HVCB) and then lead to hazard resultants. In this paper, a fault diagnosis method for the HVCB is proposed, where Gaussian prototypical networks (GPN) are adopted to realize fault detection in high accuracy only with few samples. Due to the introduction of the self-attention mechanism (SAM), the extracted fault features are more distinctive. Meanwhile, the episodic training strategy is adopted to avoid the overfitting phenomenon of the model. Finally, the effectiveness of the proposed method is verified by the experimental results.