Micro　grid　fault　of　rapid　detection　and　removal　is　the　key　to　ensure　its　reliability.　With　the　access　of　many　distributed　generations　(DG)　to　the　system,　the　characteristics　of　short-circuit　faults　of　microgrids　in　grid-connected　and　islanded　modes　have　changed,　and　the　traditional　protection　methods　can　no　longer　be　applied　to　both　operating　modes　of　microgrids　simultaneously.　Due　to　the　advantages　of　wavelet　energy　spectrum　in　the　identification　of　the　mutation　characteristics　of　the　weak　signal　as　well　as　that　of　neural　network　in　the　location　accuracy,　this　paper　proposes　a　short　circuit　fault　detection　and　protection　method　in　AC　microgrids.　The　method　takes　the　current　at　the　detection　point　as　the　object　of　analysis,　uses　the　wavelet　energy　spectrum　transform　to　analyze　the　current　waveforms　under　normal　and　fault　operation　states,　and　extracts　the　fault　characteristic　quantities.　At　the　same　time,　considering　the　effect　of　transition　resistance,　a　generalized　fault　area　identification　model　for　both　grid-connected　and　islanded　modes　is　established　by　using　a　neural　network　algorithm.　Simulation　and　experimental　results　show　that　this　method　can　realize　accurate　judgment　and　area　location　of　short-circuit　faults　in　different　modes,　different　DG　capacities,　different　fault　types　and　different　fault　regions.　IEEE