A new method is proposed to address the problem of low accuracy of existing fault line selection methods in single-phase grounding faults of small current distribution network grounding systems. Firstly, the one-dimensional zero-sequence current signal of the line is transformed into a two-dimensional image by using the Gram Angular Field. Secondly, the convolutional neural network is optimized by the grid search algorithm, and the feature map is used as the input to train and evaluate the model to achieve high-accuracy fault line selection. Experimental results show that compared with the traditional Gram Angular Field-convolutional neural network fault line selection method, this method has better line selection accuracy and provides a new idea for fault line selection in small current grounding systems.