In　order　to　improve　the　accuracy　and　efficiency　of　fault　arc　detection,　this　paper　adopts　the　characteristic　quantity　of　arc　current　wavelet　decomposition　coefficient　to　pre-judge　arc　signals,　and　adopts　the　wavelet　Mallet　algorithm　to　perform　multi-scale　decomposition　of　arc　signals.　The　decomposed　wavelet　coefficient　of　arc　current　wave　is　taken　as　the　characteristic　quantity,　and　an　adaptive　threshold　algorithm　is　proposed　for　detection.　According　to　the　results　of　the　threshold　detection　algorithm,　the　current　signal　is　selectively　input　into　the　residual　neural　network,　which　effectively　improves　the　detection　accuracy.　Finally,　the　feasibility　of　detecting　fault　arc　with　residual　neural　network　is　verified,　and　its　accuracy　rate　reaches　98.73%.　©　Beijing　Paike　Culture　Commu.　Co.,　Ltd.　2024.