A　novel　fault　classification　method　based　on　CDBN(Convolutional　Deep　Belief　Network)　for　distribution　network　is　proposed.　The　DWPT(Discrete　Wavelet　Packet　Transform)　is　adopted　to　decompose　signals　of　the　main　transformer　including　current　of　low-voltage　inlet　line,　bus　voltage,　and　so　on,　to　construct　time-frequency　matrices.　Then　the　time-frequency　matrices　are　transformed　into　the　pixel　matrices　of　the　time-frequency　spectrum　map,　which　is　used　as　the　input　of　CDBN.　Then　the　fault　features　are　autonomously　extracted　by　CDBN,　and　the　fault　classification　and　recognition　of　distribution　network　is　completed.　Fault　classification　test　is　carried　out　with　simulative　data　and　experimental　samples　of　a　typical　structure　distribution　network.　The　testing　results　show　that　the　proposed　method　not　only　can　extract　obvious　fault　characteristics　with　high　classification　accuracy,　but　also　adapts　well　to　the　change　of　neutral-point　grounding　mode　and　system　network　structure,　fault　detection　delay　and　connection　of　distributed　generation.　©　2019,　Electric　Power　Automation　Equipment　Press.　All　right　reserved.