To　solve　the　problems　of　flexible　DC　power　distribution　system,　including　flexible　operations,　multiple　fault　types,　and　difficulty　in　fault　identification,　a　fault　detection　method　based　on　K-L　divergence　optimization　variational　mode　decomposition　(VMD)　and　convolution　neural　network　(CNN)　combined　with　Inception　was　proposed.　Firstly,　K-L　VMD　method　was　used　to　extract　the　feature　component　of　the　time-domain　waveform　of　the　positive　transient　voltage　at　the　fault　point,　and　the　identification　criterion　was　constructed　using　the　feature　modal　component.　Then,　CNN　training　was　performed　on　the　sampled　data　to　obtain　the　optimal　parameters　of　the　model.　Finally,　a　10　kV　two-end　DC　distribution　network　structure　based　on　modular　multilevel　converter　(MMC)　was　built　using　the　simulation　platform　to　verify　effectiveness　of　the　proposed　method.　Simulation　experiments　show　that　K-L　divergence　optimization　variational　mode　decomposition　has　good　generalization　ability　and　anti-interference　ability　to　the　simulation　data.　The　proposed　fault　detection　method　is　effective　and　has　strong　sensitivity　to　the　identification　of　various　fault　types　and　can　accurately　identify　the　fault　types.　©　2025　Editorial　Department　of　Electric　Machines　and　Control.　All　rights　reserved.