Abstract: To address the issues of weak transient characteristics and poor positioning stability of single-phase grounding faults in distribution networks when there is high resistance grounding, arc grounding, or when the initial phase angle of the fault is small, an intelligent identification and positioning method integrating transient time-frequency features and deep temporal learning is proposed. Firstly, adaptive wavelet threshold denoising and VMD-HHT joint processing are adopted to extract multi-dimensional features such as the center frequency, energy proportion, and instantaneous frequency of the transient zero-sequence current; Secondly, a CNN-BiLSTM dual-channel fusion model is constructed, where the convolutional network is used to extract local waveform mutation features, and the bidirectional temporal memory structure is utilized to depict the transient evolution pattern; Finally, by combining waveform similarity matching and multi-measurement point energy attenuation characteristics, the fault section discrimination and location refinement are achieved. On-site recording wave data and ablation experiments results show that, the proposed method achieves a fault identification accuracy of 98.2% and a section location accuracy of 96.5% on the simulation test set, and a recognition accuracy of 95.8% and a location accuracy of 93.6% in the on-site data verification, demonstrating good noise resistance, generalization ability, and engineering application value.