Abstract: The high impedance fault (HIF) in distribution networks exhibits complex and diverse characteristics. Knowledge-driven methods face challenges in adapting to diversified fault scenarios, while data-driven methods overlook mechanism analysis in the decision-making process, leading to a lack of interpretability. To address this issue, the paper proposes a combined data-knowledge detection method for HIF. First, a qualitative analysis is conducted on the diverse distortion characteristics of zero-sequence current waveforms for HIF under different grounding dielectrics. It is concluded that the diversification of distortions manifests in terms of distortion degree, distortion offset, and irregular distortions. Then, by quantitatively analyzing the slope characteristics of volt-ampere characteristic curves under different media conditions, the applicability of knowledge-driven detection methods based on voltage-current characteristics analysis is delineated. In combination with data-driven detection methods based on support vector machines, a data-knowledge combined driving model is established based on a guided mechanism. The distortion coefficient, offset coefficient, and irregular distortion criteria are used to quantitatively describe waveform distortion characteristics. By guiding the detection model according to the specific fault scenarios, HIF detection for different fault scenarios is achieved. The simulation results show that the proposed method has an accuracy of up to 98.2% in HIF datasets containing multiple distortion types, and it can sensitively and reliably detect 8 kΩ distribution network HIFs.