In　order　to　quickly　detect　and　reliably　identify　fault　of　DC　distribution　network,　a　fault　detection　scheme　based　on　IEWT(Improved　Empirical　Wavelet　Transform)　and　IMDMF(Improved　Multi-view　Deep　Matrix　Factorization)　is　proposed.　The　local　phase-frequency　spectra　function　of　fault　current　is　fitted　nonlinearly　by　least　square　method,　based　on　which,　phase-frequency　response　of　empirical　wavelet　function　is　modified　under　certain　conditions　to　match　the　phase-frequency　spectra　characteristics　of　fault　current　as　much　as　possible.　The　IEWT　is　used　to　decompose　the　current,　and　the　modulus　maximum　of　detail　component　c3　is　calculated　to　construct　the　fault　detection　criterion.　A　weighted　self-learning　network　is　designed,　according　to　the　importance　of　the　data　to　the　classification　task,　different　weights　are　allocated　and　nested　in　the　front　of　the　multi-view　deep　matrix　factorization　model.　The　fault　features　are　extracted　from　the　current　component　c1,　c2,　and　c3,　and　the　inter　electrode　voltage　udc　by　using　the　IMDMF,　and　the　fault　classification　is　realized　by　the　soft　distribution　layer.　The　results　of　simulation　test　show　that　the　proposed　fault　detection　scheme　can　meet　the　requirements　of　speed　and　reliability　for　fault　detection,　and　the　fault　classification　accuracy　is　high,　which　lays　a　good　foundation　for　subsequent　fault　processing.　©　2022,　Electric　Power　Automation　Equipment　Press.　All　right　reserved.