In　order　to　quickly　and　reliably　detect　DC　distribution　line　faults　and　lay　a　good　foundation　for　subsequent　fault　treatment,　this　paper　proposes　a　DC　distribution　system　fault　detection　scheme　based　on　EWT　(empirical　wavelet　transform).　When　a　line　fault　occurs　in　a　DC　distribution　system,　the　high　frequency　component　of　the　line　current　is　quite　different　from　that　under　normal　conditions　and　load　switching　conditions.　Firstly,　high　frequency　mode　component　named　f2　was　obtained　from　fault　current　signal　decomposed　with　EWT.　Then　the　instantaneous　amplitude　of　f2　was　identified　by　Hilbert　transform.　After　that,　the　fault　detection　criterion　was　constructed.　An　AE　(automatic　encoder)　was　used　to　extract　the　fault　features　of　the　high　frequency　mode　component　f0~2　and　the　positive　voltage　up,　the　fault　classification　was　carried　out　by　SVM　(support　vector　machine),　and　fast　and　effective　detection　and　classification　of　DC　distribution　network　are　realized.　The　test　results　of　simulation　show　that　the　detection　scheme　can　meet　the　requirements　of　fault　detection　speed　and　reliability,　and　lay　a　good　foundation　for　subsequent　fault　processing.　©　2021,　Harbin　University　of　Science　and　Technology　Publication.　All　right　reserved.