When a high-impedance ground fault occurs in a distribution network，traditional knowledge-driven methods suffer from low accuracy in threshold selection，while data-driven methods have poor interpretability due to the lack of mechanism support.To address this problem，a data-knowledge joint-driven approach was proposed for identifying high-impedance ground faults in distribution networks.First，wavelet packet time-frequency entropy was used to quantitatively analyze the panoramic characteristics of high-impedance ground faults and normal disturbance conditions，thus revealing significant differences in their time-frequency distributions.Then，by qualitatively analyzing the characteristics of time-frequency energy spectrum matrices at different fault points，the knowledge-driven identification method based on the time-frequency energy spectrum matrix and the Transformer-based data-driven identification method were established.Guided by the characteristics of the time-frequency energy spectrum matrix，a data-knowledge joint-driven model based on a series mechanism was constructed.Finally，the simulation results of the IEEE 33-node system established in PSCAD simulation software show that the accuracy of the proposed method reaches 97.8%，and that it can accurately and sensitively detect a high-impedance ground fault with a resistance of 10 kΩ in a distribution network.